将子模块转换为普通目录

docker-hub/MapTRv2/MapTR/projects/configs/_base_/schedules/seg_cosine_150e.py

+# optimizer
+# This schedule is mainly used on S3DIS dataset in segmentation task
+optimizer = dict(type='SGD', lr=0.2, weight_decay=0.0001, momentum=0.9)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', warmup=None, min_lr=0.002)
+momentum_config = None
+
+# runtime settings
+runner = dict(type='EpochBasedRunner', max_epochs=150)

docker-hub/MapTRv2/MapTR/projects/configs/_base_/schedules/seg_cosine_200e.py

+# optimizer
+# This schedule is mainly used on ScanNet dataset in segmentation task
+optimizer = dict(type='Adam', lr=0.001, weight_decay=0.01)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', warmup=None, min_lr=1e-5)
+momentum_config = None
+
+# runtime settings
+runner = dict(type='EpochBasedRunner', max_epochs=200)

docker-hub/MapTRv2/MapTR/projects/configs/_base_/schedules/seg_cosine_50e.py

+# optimizer
+# This schedule is mainly used on S3DIS dataset in segmentation task
+optimizer = dict(type='Adam', lr=0.001, weight_decay=0.001)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(policy='CosineAnnealing', warmup=None, min_lr=1e-5)
+momentum_config = None
+
+# runtime settings
+runner = dict(type='EpochBasedRunner', max_epochs=50)

docker-hub/MapTRv2/MapTR/projects/configs/bevformer/bevformer_base.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+voxel_size = [0.2, 0.2, 8]
+
+
+
+img_norm_cfg = dict(
+    mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 4
+bev_h_ = 200
+bev_w_ = 200
+queue_length = 4 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='BEVFormer',
+    use_grid_mask=True,
+    video_test_mode=True,
+    img_backbone=dict(
+        type='ResNet',
+        depth=101,
+        num_stages=4,
+        out_indices=(1, 2, 3),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN2d', requires_grad=False),
+        norm_eval=True,
+        style='caffe',
+        dcn=dict(type='DCNv2', deform_groups=1, fallback_on_stride=False), # original DCNv2 will print log when perform load_state_dict
+        stage_with_dcn=(False, False, True, True)),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[512, 1024, 2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=4,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='BEVFormerHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_classes=10,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        transformer=dict(
+            type='PerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=6,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='SpatialCrossAttention',
+                            pc_range=point_cloud_range,
+                            deformable_attention=dict(
+                                type='MSDeformableAttention3D',
+                                embed_dims=_dim_,
+                                num_points=8,
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='DetectionTransformerDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='NMSFreeCoder',
+            post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            pc_range=point_cloud_range,
+            max_num=300,
+            voxel_size=voxel_size,
+            num_classes=10),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.25),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='HungarianAssigner3D',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            iou_cost=dict(type='IoUCost', weight=0.0), # Fake cost. This is just to make it compatible with DETR head.
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=1,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=2e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+evaluation = dict(interval=1, pipeline=test_pipeline)
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+load_from = 'ckpts/r101_dcn_fcos3d_pretrain.pth'
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+
+checkpoint_config = dict(interval=1)

docker-hub/MapTRv2/MapTR/projects/configs/bevformer/bevformer_small.py

+# BEvFormer-small consumes at lease 10500M GPU memory
+# compared to bevformer_base, bevformer_small has
+# smaller BEV: 200*200 -> 150*150
+# less encoder layers: 6 -> 3
+# smaller input size: 1600*900 -> (1600*900)*0.8
+# multi-scale feautres -> single scale features (C5)
+# with_cp of backbone = True
+
+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+voxel_size = [0.2, 0.2, 8]
+
+
+img_norm_cfg = dict(
+    mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+bev_h_ = 150
+bev_w_ = 150
+queue_length = 3 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='BEVFormer',
+    use_grid_mask=True,
+    video_test_mode=True,
+    img_backbone=dict(
+        type='ResNet',
+        depth=101,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN2d', requires_grad=False),
+        norm_eval=True,
+        style='caffe',
+        with_cp=True, # using checkpoint to save GPU memory
+        dcn=dict(type='DCNv2', deform_groups=1, fallback_on_stride=False), # original DCNv2 will print log when perform load_state_dict
+        stage_with_dcn=(False, False, True, True)),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='BEVFormerHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_classes=10,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        transformer=dict(
+            type='PerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=3,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='SpatialCrossAttention',
+                            pc_range=point_cloud_range,
+                            deformable_attention=dict(
+                                type='MSDeformableAttention3D',
+                                embed_dims=_dim_,
+                                num_points=8,
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='DetectionTransformerDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='NMSFreeCoder',
+            post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            pc_range=point_cloud_range,
+            max_num=300,
+            voxel_size=voxel_size,
+            num_classes=10),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.25),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='HungarianAssigner3D',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            iou_cost=dict(type='IoUCost', weight=0.0), # Fake cost. This is just to make it compatible with DETR head.
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.8]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    # dict(type='PadMultiViewImage', size_divisor=32),
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.8]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=1,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=2e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+evaluation = dict(interval=1, pipeline=test_pipeline)
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+load_from = 'ckpts/r101_dcn_fcos3d_pretrain.pth'
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+
+checkpoint_config = dict(interval=1)

docker-hub/MapTRv2/MapTR/projects/configs/bevformer/bevformer_tiny.py

+# BEvFormer-tiny consumes at lease 6700M GPU memory
+# compared to bevformer_base, bevformer_tiny has
+# smaller backbone: R101-DCN -> R50
+# smaller BEV: 200*200 -> 50*50
+# less encoder layers: 6 -> 3
+# smaller input size: 1600*900 -> 800*450
+# multi-scale feautres -> single scale features (C5)
+
+
+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+voxel_size = [0.2, 0.2, 8]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+bev_h_ = 50
+bev_w_ = 50
+queue_length = 3 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='BEVFormer',
+    use_grid_mask=True,
+    video_test_mode=True,
+    pretrained=dict(img='torchvision://resnet50'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='BEVFormerHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_classes=10,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        transformer=dict(
+            type='PerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=3,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='SpatialCrossAttention',
+                            pc_range=point_cloud_range,
+                            deformable_attention=dict(
+                                type='MSDeformableAttention3D',
+                                embed_dims=_dim_,
+                                num_points=8,
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='DetectionTransformerDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='NMSFreeCoder',
+            post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            pc_range=point_cloud_range,
+            max_num=300,
+            voxel_size=voxel_size,
+            num_classes=10),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.25),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='HungarianAssigner3D',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            iou_cost=dict(type='IoUCost', weight=0.0), # Fake cost. This is just to make it compatible with DETR head.
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=1,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=2e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+evaluation = dict(interval=1, pipeline=test_pipeline)
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+
+checkpoint_config = dict(interval=1)

docker-hub/MapTRv2/MapTR/projects/configs/bevformer_fp16/bevformer_tiny_fp16.py

+# BEvFormer-tiny consumes at lease 6700M GPU memory
+# compared to bevformer_base, bevformer_tiny has
+# smaller backbone: R101-DCN -> R50
+# smaller BEV: 200*200 -> 50*50
+# less encoder layers: 6 -> 3
+# smaller input size: 1600*900 -> 800*450
+# multi-scale feautres -> single scale features (C5)
+
+
+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+voxel_size = [0.2, 0.2, 8]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+bev_h_ = 50
+bev_w_ = 50
+queue_length = 3 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='BEVFormer_fp16',
+    use_grid_mask=True,
+    video_test_mode=True,
+    pretrained=dict(img='torchvision://resnet50'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='BEVFormerHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_classes=10,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        transformer=dict(
+            type='PerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=3,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='SpatialCrossAttention',
+                            pc_range=point_cloud_range,
+                            deformable_attention=dict(
+                                type='MSDeformableAttention3D',
+                                embed_dims=_dim_,
+                                num_points=8,
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='DetectionTransformerDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='NMSFreeCoder',
+            post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            pc_range=point_cloud_range,
+            max_num=300,
+            voxel_size=voxel_size,
+            num_classes=10),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.25),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='HungarianAssigner3D',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            iou_cost=dict(type='IoUCost', weight=0.0), # Fake cost. This is just to make it compatible with DETR head.
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=2,
+    workers_per_gpu=8,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=2.8e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+evaluation = dict(interval=1, pipeline=test_pipeline)
+
+runner = dict(type='EpochBasedRunner_video', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)
+custom_hooks = [dict(type='TransferWeight',priority='LOWEST')]
\ No newline at end of file

docker-hub/MapTRv2/MapTR/projects/configs/datasets/custom_lyft-3d.py

+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+point_cloud_range = [-80, -80, -5, 80, 80, 3]
+# For Lyft we usually do 9-class detection
+class_names = [
+    'car', 'truck', 'bus', 'emergency_vehicle', 'other_vehicle', 'motorcycle',
+    'bicycle', 'pedestrian', 'animal'
+]
+dataset_type = 'CustomLyftDataset'
+data_root = 'data/lyft/'
+# Input modality for Lyft dataset, this is consistent with the submission
+# format which requires the information in input_modality.
+input_modality = dict(
+    use_lidar=True,
+    use_camera=False,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+file_client_args = dict(backend='disk')
+# Uncomment the following if use ceph or other file clients.
+# See https://mmcv.readthedocs.io/en/latest/api.html#mmcv.fileio.FileClient
+# for more details.
+# file_client_args = dict(
+#     backend='petrel',
+#     path_mapping=dict({
+#         './data/lyft/': 's3://lyft/lyft/',
+#         'data/lyft/': 's3://lyft/lyft/'
+#    }))
+train_pipeline = [
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        file_client_args=file_client_args),
+    dict(
+        type='LoadPointsFromMultiSweeps',
+        sweeps_num=10,
+        file_client_args=file_client_args),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
+    dict(
+        type='GlobalRotScaleTrans',
+        rot_range=[-0.3925, 0.3925],
+        scale_ratio_range=[0.95, 1.05],
+        translation_std=[0, 0, 0]),
+    dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5),
+    dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='PointShuffle'),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
+]
+test_pipeline = [
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        file_client_args=file_client_args),
+    dict(
+        type='LoadPointsFromMultiSweeps',
+        sweeps_num=10,
+        file_client_args=file_client_args),
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1333, 800),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(
+                type='GlobalRotScaleTrans',
+                rot_range=[0, 0],
+                scale_ratio_range=[1., 1.],
+                translation_std=[0, 0, 0]),
+            dict(type='RandomFlip3D'),
+            dict(
+                type='PointsRangeFilter', point_cloud_range=point_cloud_range),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='Collect3D', keys=['points'])
+        ])
+]
+# construct a pipeline for data and gt loading in show function
+# please keep its loading function consistent with test_pipeline (e.g. client)
+eval_pipeline = [
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        file_client_args=file_client_args),
+    dict(
+        type='LoadPointsFromMultiSweeps',
+        sweeps_num=10,
+        file_client_args=file_client_args),
+    dict(
+        type='DefaultFormatBundle3D',
+        class_names=class_names,
+        with_label=False),
+    dict(type='Collect3D', keys=['points'])
+]
+
+data = dict(
+    samples_per_gpu=2,
+    workers_per_gpu=2,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'lyft_infos_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False),
+    val=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'lyft_infos_val.pkl',
+        pipeline=test_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=True),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'lyft_infos_val.pkl',
+        pipeline=test_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=True))
+# For Lyft dataset, we usually evaluate the model at the end of training.
+# Since the models are trained by 24 epochs by default, we set evaluation
+# interval to be 24. Please change the interval accordingly if you do not
+# use a default schedule.
+evaluation = dict(interval=24, pipeline=eval_pipeline)
\ No newline at end of file

docker-hub/MapTRv2/MapTR/projects/configs/datasets/custom_nus-3d.py

+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+point_cloud_range = [-50, -50, -5, 50, 50, 3]
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
+    'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'
+]
+dataset_type = 'NuScenesDataset_eval_modified'
+data_root = 'data/nuscenes/'
+# Input modality for nuScenes dataset, this is consistent with the submission
+# format which requires the information in input_modality.
+input_modality = dict(
+    use_lidar=True,
+    use_camera=False,
+    use_radar=False,
+    use_map=False,
+    use_external=False)
+file_client_args = dict(backend='disk')
+# Uncomment the following if use ceph or other file clients.
+# See https://mmcv.readthedocs.io/en/latest/api.html#mmcv.fileio.FileClient
+# for more details.
+# file_client_args = dict(
+#     backend='petrel',
+#     path_mapping=dict({
+#         './data/nuscenes/': 's3://nuscenes/nuscenes/',
+#         'data/nuscenes/': 's3://nuscenes/nuscenes/'
+#     }))
+train_pipeline = [
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        file_client_args=file_client_args),
+    dict(
+        type='LoadPointsFromMultiSweeps',
+        sweeps_num=10,
+        file_client_args=file_client_args),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
+    dict(
+        type='GlobalRotScaleTrans',
+        rot_range=[-0.3925, 0.3925],
+        scale_ratio_range=[0.95, 1.05],
+        translation_std=[0, 0, 0]),
+    dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5),
+    dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='PointShuffle'),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
+]
+test_pipeline = [
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        file_client_args=file_client_args),
+    dict(
+        type='LoadPointsFromMultiSweeps',
+        sweeps_num=10,
+        file_client_args=file_client_args),
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1333, 800),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(
+                type='GlobalRotScaleTrans',
+                rot_range=[0, 0],
+                scale_ratio_range=[1., 1.],
+                translation_std=[0, 0, 0]),
+            dict(type='RandomFlip3D'),
+            dict(
+                type='PointsRangeFilter', point_cloud_range=point_cloud_range),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='Collect3D', keys=['points'])
+        ])
+]
+# construct a pipeline for data and gt loading in show function
+# please keep its loading function consistent with test_pipeline (e.g. client)
+eval_pipeline = [
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        file_client_args=file_client_args),
+    dict(
+        type='LoadPointsFromMultiSweeps',
+        sweeps_num=10,
+        file_client_args=file_client_args),
+    dict(
+        type='DefaultFormatBundle3D',
+        class_names=class_names,
+        with_label=False),
+    dict(type='Collect3D', keys=['points'])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'nuscenes_infos_val.pkl',
+        pipeline=test_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=True,
+        box_type_3d='LiDAR'),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_val.pkl',
+        pipeline=test_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=True,
+        box_type_3d='LiDAR'))
+# For nuScenes dataset, we usually evaluate the model at the end of training.
+# Since the models are trained by 24 epochs by default, we set evaluation
+# interval to be 24. Please change the interval accordingly if you do not
+# use a default schedule.
+evaluation = dict(interval=24, pipeline=eval_pipeline)

docker-hub/MapTRv2/MapTR/projects/configs/datasets/custom_waymo-3d.py

+# dataset settings
+# D5 in the config name means the whole dataset is divided into 5 folds
+# We only use one fold for efficient experiments
+dataset_type = 'CustomWaymoDataset'
+data_root = 'data/waymo/kitti_format/'
+file_client_args = dict(backend='disk')
+# Uncomment the following if use ceph or other file clients.
+# See https://mmcv.readthedocs.io/en/latest/api.html#mmcv.fileio.FileClient
+# for more details.
+# file_client_args = dict(
+#     backend='petrel', path_mapping=dict(data='s3://waymo_data/'))
+
+img_norm_cfg = dict(
+    mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
+class_names = ['Car', 'Pedestrian', 'Cyclist']
+point_cloud_range = [-74.88, -74.88, -2, 74.88, 74.88, 4]
+input_modality = dict(use_lidar=False, use_camera=True)
+db_sampler = dict(
+    data_root=data_root,
+    info_path=data_root + 'waymo_dbinfos_train.pkl',
+    rate=1.0,
+    prepare=dict(
+        filter_by_difficulty=[-1],
+        filter_by_min_points=dict(Car=5, Pedestrian=10, Cyclist=10)),
+    classes=class_names,
+    sample_groups=dict(Car=15, Pedestrian=10, Cyclist=10),
+    points_loader=dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=[0, 1, 2, 3, 4],
+        file_client_args=file_client_args))
+
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1920, 1280),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+
+# construct a pipeline for data and gt loading in show function
+# please keep its loading function consistent with test_pipeline (e.g. client)
+
+data = dict(
+    samples_per_gpu=2,
+    workers_per_gpu=4,
+    train=dict(
+        type='RepeatDataset',
+        times=2,
+        dataset=dict(
+            type=dataset_type,
+            data_root=data_root,
+            ann_file=data_root + 'waymo_infos_train.pkl',
+            split='training',
+            pipeline=train_pipeline,
+            modality=input_modality,
+            classes=class_names,
+            test_mode=False,
+            # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+            # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+            box_type_3d='LiDAR',
+            # load one frame every five frames
+            load_interval=5)),
+    val=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'waymo_infos_val.pkl',
+        split='training',
+        pipeline=test_pipeline,
+        modality=input_modality,
+        classes=class_names,
+        test_mode=True,
+        box_type_3d='LiDAR'),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'waymo_infos_val.pkl',
+        split='training',
+        pipeline=test_pipeline,
+        modality=input_modality,
+        classes=class_names,
+        test_mode=True,
+        box_type_3d='LiDAR'))
+
+evaluation = dict(interval=24, pipeline=test_pipeline)
\ No newline at end of file

docker-hub/MapTRv2/MapTR/projects/configs/maptr/maptr_nano_r18_110e.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-15.0, -30.0, -2.0, 15.0, 30.0, 2.0]
+voxel_size = [0.15, 0.15, 4]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 80
+bev_w_ = 40
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='MapTR',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet18-f37072fd.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=18,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=-1,
+        norm_cfg=dict(type='SyncBN', requires_grad=True),
+        norm_eval=False,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[512],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec=100,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=1,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='GeometrySptialCrossAttention',
+                            pc_range=point_cloud_range,
+                            attention=dict(
+                                type='GeometryKernelAttention',
+                                embed_dims=_dim_,
+                                num_heads=4,
+                                dilation=1,
+                                kernel_size=(3,5),
+                                num_levels=_num_levels_,
+                                im2col_step=192),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=2,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=4,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1,
+                            im2col_step=192),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-20, -35, -20, -35, 20, 35, 20, 35],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesLocalMapDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.2]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.2]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=24,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             pc_range=point_cloud_range,
+             fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+             eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+             padding_value=-10000,
+             map_classes=map_classes,
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              pc_range=point_cloud_range,
+              fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+              eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+              padding_value=-10000,
+              map_classes=map_classes,
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=4e-3,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=50, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 110
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer')
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=5)

docker-hub/MapTRv2/MapTR/projects/configs/maptr/maptr_tiny_fusion_24e.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-15.0, -30.0, -2.0, 15.0, 30.0, 2.0]
+lidar_point_cloud_range = [-15.0, -30.0, -5.0, 15.0, 30.0, 3.0]
+voxel_size = [0.1, 0.1, 0.2]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=True,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 200
+bev_w_ = 100
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='MapTR',
+    use_grid_mask=True,
+    video_test_mode=False,
+    modality='fusion',
+    lidar_encoder=dict(
+        voxelize=dict(max_num_points=10,point_cloud_range=lidar_point_cloud_range,voxel_size=voxel_size,max_voxels=[90000, 120000]),
+        backbone=dict(
+            type='SparseEncoder',
+            in_channels=5,
+            sparse_shape=[300, 600, 41],
+            output_channels=128,
+            order=('conv', 'norm', 'act'),
+            encoder_channels=((16, 16, 32), (32, 32, 64), (64, 64, 128), (128,
+                                                                        128)),
+            encoder_paddings=([0, 0, 1], [0, 0, 1], [0, 0, [1, 1, 0]], [0, 0]),
+            block_type='basicblock'
+        ),
+    ),
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec=50,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            modality='fusion',
+            fuser=dict(
+                type='ConvFuser',
+                in_channels=[_dim_, 256],
+                out_channels=_dim_,
+            ),
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=1,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='GeometrySptialCrossAttention',
+                            pc_range=point_cloud_range,
+                            attention=dict(
+                                type='GeometryKernelAttention',
+                                embed_dims=_dim_,
+                                num_heads=4,
+                                dilation=1,
+                                kernel_size=(3,5),
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-20, -35, -20, -35, 20, 35, 20, 35],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesLocalMapDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+reduce_beams=32
+load_dim=5
+use_dim=5
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='CustomLoadPointsFromFile', coord_type='LIDAR', load_dim=load_dim, use_dim=use_dim, reduce_beams=reduce_beams),
+    dict(type='CustomLoadPointsFromMultiSweeps', sweeps_num=9, load_dim=load_dim, use_dim=use_dim, reduce_beams=reduce_beams, pad_empty_sweeps=True, remove_close=True),
+    dict(type='CustomPointsRangeFilter', point_cloud_range=lidar_point_cloud_range),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img', 'points'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='CustomLoadPointsFromFile', coord_type='LIDAR', load_dim=load_dim, use_dim=use_dim, reduce_beams=reduce_beams),
+    dict(type='CustomLoadPointsFromMultiSweeps', sweeps_num=9, load_dim=load_dim, use_dim=use_dim, reduce_beams=reduce_beams, pad_empty_sweeps=True, remove_close=True),
+    dict(type='CustomPointsRangeFilter', point_cloud_range=lidar_point_cloud_range),
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img', 'points'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             pc_range=point_cloud_range,
+             fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+             eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+             padding_value=-10000,
+             map_classes=map_classes,
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              pc_range=point_cloud_range,
+              fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+              eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+              padding_value=-10000,
+              map_classes=map_classes,
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer')
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)
+find_unused_parameters=True
+

docker-hub/MapTRv2/MapTR/projects/configs/maptr/maptr_tiny_r50_110e.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-15.0, -30.0, -2.0, 15.0, 30.0, 2.0]
+voxel_size = [0.15, 0.15, 4]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 200
+bev_w_ = 100
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='MapTR',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec=50,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=1,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='GeometrySptialCrossAttention',
+                            pc_range=point_cloud_range,
+                            attention=dict(
+                                type='GeometryKernelAttention',
+                                embed_dims=_dim_,
+                                num_heads=4,
+                                dilation=1,
+                                kernel_size=(3,5),
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-20, -35, -20, -35, 20, 35, 20, 35],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesLocalMapDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             pc_range=point_cloud_range,
+             fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+             eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+             padding_value=-10000,
+             map_classes=map_classes,
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              pc_range=point_cloud_range,
+              fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+              eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+              padding_value=-10000,
+              map_classes=map_classes,
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 110
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer')
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=5)

docker-hub/MapTRv2/MapTR/projects/configs/maptr/maptr_tiny_r50_24e.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-15.0, -30.0, -2.0, 15.0, 30.0, 2.0]
+voxel_size = [0.15, 0.15, 4]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 200
+bev_w_ = 100
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='MapTR',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec=50,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=1,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='GeometrySptialCrossAttention',
+                            pc_range=point_cloud_range,
+                            attention=dict(
+                                type='GeometryKernelAttention',
+                                embed_dims=_dim_,
+                                num_heads=4,
+                                dilation=1,
+                                kernel_size=(3,5),
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-20, -35, -20, -35, 20, 35, 20, 35],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesLocalMapDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             pc_range=point_cloud_range,
+             fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+             eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+             padding_value=-10000,
+             map_classes=map_classes,
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              pc_range=point_cloud_range,
+              fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+              eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+              padding_value=-10000,
+              map_classes=map_classes,
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer')
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)

docker-hub/MapTRv2/MapTR/projects/configs/maptr/maptr_tiny_r50_24e_bevformer.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-15.0, -30.0, -2.0, 15.0, 30.0, 2.0]
+voxel_size = [0.15, 0.15, 4]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 200
+bev_w_ = 100
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='MapTR',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec=50,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=1,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='SpatialCrossAttention',
+                            pc_range=point_cloud_range,
+                            deformable_attention=dict(
+                                type='MSDeformableAttention3D',
+                                embed_dims=_dim_,
+                                num_points=8,
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-20, -35, -20, -35, 20, 35, 20, 35],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesLocalMapDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             pc_range=point_cloud_range,
+             fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+             eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+             padding_value=-10000,
+             map_classes=map_classes,
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              pc_range=point_cloud_range,
+              fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+              eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+              padding_value=-10000,
+              map_classes=map_classes,
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer')
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)

docker-hub/MapTRv2/MapTR/projects/configs/maptr/maptr_tiny_r50_24e_bevpool.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-15.0, -30.0,-10.0, 15.0, 30.0, 10.0]
+voxel_size = [0.15, 0.15, 20.0]
+dbound=[1.0, 35.0, 0.5]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 200
+bev_w_ = 100
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='MapTR',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec=50,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='LSSTransform',
+                in_channels=_dim_,
+                out_channels=_dim_,
+                feat_down_sample=32,
+                pc_range=point_cloud_range,
+                voxel_size=voxel_size,
+                dbound=dbound,
+                downsample=2),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-20, -35, -20, -35, 20, 35, 20, 35],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesLocalMapDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True, with_attr_label=False),
+    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
+    dict(type='ObjectNameFilter', classes=class_names),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['gt_bboxes_3d', 'gt_labels_3d', 'img'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+             map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             pc_range=point_cloud_range,
+             fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+             eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+             padding_value=-10000,
+             map_classes=map_classes,
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'nuscenes_infos_temporal_val.pkl',
+              map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              pc_range=point_cloud_range,
+              fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+              eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+              padding_value=-10000,
+              map_classes=map_classes,
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer')
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)

docker-hub/MapTRv2/MapTR/projects/configs/maptr/maptr_tiny_r50_av2_24e.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-30.0, -15.0, -2.0, 30.0, 15.0, 2.0]
+voxel_size = [0.15, 0.15, 4]
+
+
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 100
+bev_w_ = 200
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+model = dict(
+    type='MapTR',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRHead',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec=50,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            num_cams=7,
+            encoder=dict(
+                type='BEVFormerEncoder',
+                num_layers=1,
+                pc_range=point_cloud_range,
+                num_points_in_pillar=4,
+                return_intermediate=False,
+                transformerlayers=dict(
+                    type='BEVFormerLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='TemporalSelfAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                        dict(
+                            type='GeometrySptialCrossAttention',
+                            pc_range=point_cloud_range,
+                            num_cams=7,
+                            attention=dict(
+                                type='GeometryKernelAttention',
+                                embed_dims=_dim_,
+                                num_heads=4,
+                                dilation=1,
+                                kernel_size=(3,5),
+                                num_levels=_num_levels_),
+                            embed_dims=_dim_,
+                        )
+                    ],
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm'))),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DetrTransformerDecoderLayer',
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-35, -20, -35, -20, 35, 20, 35, 20],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005)),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomAV2LocalMapDataset'
+data_root = 'data/argoverse2/sensor/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='CustomLoadMultiViewImageFromFiles', to_float32=True, padding=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.3]),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', class_names=class_names),
+    dict(type='CustomCollect3D', keys=['img'])
+]
+
+test_pipeline = [
+    dict(type='CustomLoadMultiViewImageFromFiles', to_float32=True, padding=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.3]),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(2048, 2048), # 2048*0.3, 2048*0.3
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D',
+                class_names=class_names,
+                with_label=False),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=6,
+    workers_per_gpu=8,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'av2_map_infos_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(type=dataset_type,
+             data_root=data_root,
+             ann_file=data_root + 'av2_map_infos_val.pkl',
+            #  ann_file=data_root + 'av2_map_infos_train.pkl',
+             map_ann_file=data_root + 'av2_map_anns_val.json',
+             load_interval=4, # av2 uses 10 Hz, set to 5, 2HZ the same as nuscenes,
+            #  load_interval=1, # TODO debug
+             pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+             pc_range=point_cloud_range,
+             fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+             eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+             padding_value=-10000,
+             map_classes=map_classes,
+             classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(type=dataset_type,
+              data_root=data_root,
+              ann_file=data_root + 'av2_map_infos_val.pkl',
+            #   ann_file=data_root + 'av2_map_infos_train.pkl',
+              map_ann_file=data_root + 'av2_map_anns_val.json',
+              load_interval=4, # av2 uses 10 Hz, set to 5, 2HZ the same as nuscenes,
+            #   load_interval=1, # TODO debug
+              pipeline=test_pipeline, bev_size=(bev_h_, bev_w_),
+              pc_range=point_cloud_range,
+              fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+              eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+              padding_value=-10000,
+              map_classes=map_classes,
+              classes=class_names, modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 24
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer')
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)

docker-hub/MapTRv2/MapTR/projects/configs/maptrv2/maptrv2_av2_3d_r50_6ep.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-30.0, -15.0, -5.0, 30.0, 15.0, 3.0]
+voxel_size = [0.15, 0.15, 8.0]
+dbound=[1.0, 35.0, 0.5]
+
+grid_config = {
+    'x': [-30.0, -30.0, 0.15], # useless
+    'y': [-15.0, -15.0, 0.15], # useless
+    'z': [-10, 10, 20],        # useless
+    'depth': [1.0, 35.0, 0.5], # useful
+}
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+num_vec=50
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 100
+bev_w_ = 200
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+aux_seg_cfg = dict(
+    use_aux_seg=True,
+    bev_seg=True,
+    pv_seg=True,
+    seg_classes=1,
+    feat_down_sample=32,
+    pv_thickness=1,
+)
+
+z_cfg = dict(
+    pred_z_flag=True,
+    gt_z_flag=True,
+)
+
+model = dict(
+    type='MapTRv2',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRv2Head',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec_one2one=num_vec,
+        num_vec_one2many=300,
+        k_one2many=6,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=3,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        aux_seg=aux_seg_cfg,
+        z_cfg=z_cfg,
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            num_cams=7,
+            z_cfg=z_cfg,
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='LSSTransform',
+                in_channels=_dim_,
+                out_channels=_dim_,
+                feat_down_sample=32,
+                pc_range=point_cloud_range,
+                voxel_size=voxel_size,
+                dbound=dbound,
+                downsample=2,
+                loss_depth_weight=3.0,
+                depthnet_cfg=dict(use_dcn=False, with_cp=False, aspp_mid_channels=96),
+                grid_config=grid_config,),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DecoupledDetrTransformerDecoderLayer',
+                    num_vec=num_vec,
+                    num_pts_per_vec=fixed_ptsnum_per_pred_line,
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'self_attn', 'norm','cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            z_cfg=z_cfg,
+            post_center_range=[-35, -20, -35, -20, 35, 20, 35, 20],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005),
+        loss_seg=dict(type='SimpleLoss', 
+            pos_weight=4.0,
+            loss_weight=1.0),
+        loss_pv_seg=dict(type='SimpleLoss', 
+                    pos_weight=1.0,
+                    loss_weight=2.0),),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            z_cfg=z_cfg,
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomAV2OfflineLocalMapDataset'
+data_root = 'data/argoverse2/sensor/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='CustomLoadMultiViewImageFromFiles', to_float32=True, padding=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.3]),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', with_gt=False, with_label=False,class_names=map_classes),
+    dict(type='CustomCollect3D', keys=['img'])
+]
+
+test_pipeline = [
+    dict(type='CustomLoadMultiViewImageFromFiles', to_float32=True, padding=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.3]),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(2048, 2048), # 2048*0.3, 2048*0.3
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D', 
+                with_gt=False, 
+                with_label=False,
+                class_names=map_classes),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'av2_map_infos_train.pkl',
+        z_cfg=z_cfg,
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        aux_seg=aux_seg_cfg,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'av2_map_infos_val.pkl',
+        map_ann_file=data_root + 'av2_gt_map_anns_val.json',
+        # code_size=3,
+        z_cfg=z_cfg,
+        load_interval=4, # av2 uses 10 Hz, set to 5, 2HZ the same as nuscenes,
+        # load_interval=1, # TODO debug
+        pipeline=test_pipeline,  
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        classes=class_names, 
+        modality=input_modality, 
+        samples_per_gpu=1),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'av2_map_infos_val.pkl',
+        map_ann_file=data_root + 'av2_gt_map_anns_val.json',
+        # code_size=3,
+        z_cfg=z_cfg,
+        load_interval=4, # av2 uses 10 Hz, set to 5, 2HZ the same as nuscenes,
+        # load_interval=1, # TODO debug
+        pipeline=test_pipeline, 
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        classes=class_names, 
+        modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=200,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 6
+evaluation = dict(interval=1, pipeline=test_pipeline, metric='chamfer')
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)
+find_unused_parameters=True
\ No newline at end of file

docker-hub/MapTRv2/MapTR/projects/configs/maptrv2/maptrv2_av2_3d_r50_6ep_w_centerline.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-30.0, -15.0, -5.0, 30.0, 15.0, 3.0]
+voxel_size = [0.15, 0.15, 8.0]
+dbound=[1.0, 35.0, 0.5]
+
+grid_config = {
+    'x': [-30.0, -30.0, 0.15], # useless
+    'y': [-15.0, -15.0, 0.15], # useless
+    'z': [-10, 10, 20],        # useless
+    'depth': [1.0, 35.0, 0.5], # useful
+}
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary', 'centerline']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+num_vec=70
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 100
+bev_w_ = 200
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+aux_seg_cfg = dict(
+    use_aux_seg=True,
+    bev_seg=True,
+    pv_seg=True,
+    seg_classes=1,
+    feat_down_sample=32,
+    pv_thickness=1,
+)
+
+z_cfg = dict(
+    pred_z_flag=True,
+    gt_z_flag=True,
+)
+
+model = dict(
+    type='MapTRv2',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRv2Head',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec_one2one=num_vec,
+        num_vec_one2many=300,
+        k_one2many=6,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=3,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        aux_seg=aux_seg_cfg,
+        z_cfg=z_cfg,
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            num_cams=7,
+            z_cfg=z_cfg,
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='LSSTransform',
+                in_channels=_dim_,
+                out_channels=_dim_,
+                feat_down_sample=32,
+                pc_range=point_cloud_range,
+                voxel_size=voxel_size,
+                dbound=dbound,
+                downsample=2,
+                loss_depth_weight=3.0,
+                depthnet_cfg=dict(use_dcn=False, with_cp=False, aspp_mid_channels=96),
+                grid_config=grid_config,),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DecoupledDetrTransformerDecoderLayer',
+                    num_vec=num_vec,
+                    num_pts_per_vec=fixed_ptsnum_per_pred_line,
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'self_attn', 'norm','cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            z_cfg=z_cfg,
+            post_center_range=[-35, -20, -35, -20, 35, 20, 35, 20],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005),
+        loss_seg=dict(type='SimpleLoss', 
+            pos_weight=4.0,
+            loss_weight=1.0),
+        loss_pv_seg=dict(type='SimpleLoss', 
+                    pos_weight=1.0,
+                    loss_weight=2.0),),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            z_cfg=z_cfg,
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomAV2OfflineLocalMapDataset'
+data_root = 'data/argoverse2/sensor/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='CustomLoadMultiViewImageFromFiles', to_float32=True, padding=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.3]),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(type='PadMultiViewImage', size_divisor=32),
+    dict(type='DefaultFormatBundle3D', with_gt=False, with_label=False,class_names=map_classes),
+    dict(type='CustomCollect3D', keys=['img'])
+]
+
+test_pipeline = [
+    dict(type='CustomLoadMultiViewImageFromFiles', to_float32=True, padding=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.3]),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(2048, 2048), # 2048*0.3, 2048*0.3
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D', 
+                with_gt=False, 
+                with_label=False,
+                class_names=map_classes),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=4,
+    workers_per_gpu=4,
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'av2_map_infos_train.pkl',
+        z_cfg=z_cfg,
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        aux_seg=aux_seg_cfg,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'av2_map_infos_val.pkl',
+        map_ann_file=data_root + 'av2_gt_map_anns_val.json',
+        # code_size=3,
+        z_cfg=z_cfg,
+        load_interval=4, # av2 uses 10 Hz, set to 5, 2HZ the same as nuscenes,
+        # load_interval=1, # TODO debug
+        pipeline=test_pipeline,  
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        classes=class_names, 
+        modality=input_modality, 
+        samples_per_gpu=1),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'av2_map_infos_val.pkl',
+        map_ann_file=data_root + 'av2_gt_map_anns_val.json',
+        # code_size=3,
+        z_cfg=z_cfg,
+        load_interval=4, # av2 uses 10 Hz, set to 5, 2HZ the same as nuscenes,
+        # load_interval=1, # TODO debug
+        pipeline=test_pipeline, 
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        classes=class_names, 
+        modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=200,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 6
+evaluation = dict(interval=1, pipeline=test_pipeline, metric='chamfer')
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=50,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(interval=1)
+find_unused_parameters=True
\ No newline at end of file

docker-hub/MapTRv2/MapTR/projects/configs/maptrv2/maptrv2_nusc_r50_24ep.py

+_base_ = [
+    '../datasets/custom_nus-3d.py',
+    '../_base_/default_runtime.py'
+]
+#
+plugin = True
+plugin_dir = 'projects/mmdet3d_plugin/'
+
+# If point cloud range is changed, the models should also change their point
+# cloud range accordingly
+# point_cloud_range = [-51.2, -51.2, -5.0, 51.2, 51.2, 3.0]
+point_cloud_range = [-15.0, -30.0,-10.0, 15.0, 30.0, 10.0]
+voxel_size = [0.15, 0.15, 20.0]
+dbound=[1.0, 35.0, 0.5]
+
+grid_config = {
+    'x': [-30.0, -30.0, 0.15], # useless
+    'y': [-15.0, -15.0, 0.15], # useless
+    'z': [-10, 10, 20],        # useless
+    'depth': [1.0, 35.0, 0.5], # useful
+}
+
+
+img_norm_cfg = dict(
+    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+
+# For nuScenes we usually do 10-class detection
+class_names = [
+    'car', 'truck', 'construction_vehicle', 'bus', 'trailer', 'barrier',
+    'motorcycle', 'bicycle', 'pedestrian', 'traffic_cone'
+]
+# map has classes: divider, ped_crossing, boundary
+map_classes = ['divider', 'ped_crossing','boundary']
+# fixed_ptsnum_per_line = 20
+# map_classes = ['divider',]
+num_vec=50
+fixed_ptsnum_per_gt_line = 20 # now only support fixed_pts > 0
+fixed_ptsnum_per_pred_line = 20
+eval_use_same_gt_sample_num_flag=True
+num_map_classes = len(map_classes)
+
+input_modality = dict(
+    use_lidar=False,
+    use_camera=True,
+    use_radar=False,
+    use_map=False,
+    use_external=True)
+
+_dim_ = 256
+_pos_dim_ = _dim_//2
+_ffn_dim_ = _dim_*2
+_num_levels_ = 1
+# bev_h_ = 50
+# bev_w_ = 50
+bev_h_ = 200
+bev_w_ = 100
+queue_length = 1 # each sequence contains `queue_length` frames.
+
+aux_seg_cfg = dict(
+    use_aux_seg=True,
+    bev_seg=True,
+    pv_seg=True,
+    seg_classes=1,
+    feat_down_sample=32,
+    pv_thickness=1,
+)
+
+model = dict(
+    type='MapTRv2',
+    use_grid_mask=True,
+    video_test_mode=False,
+    pretrained=dict(img='ckpts/resnet50-19c8e357.pth'),
+    img_backbone=dict(
+        type='ResNet',
+        depth=50,
+        num_stages=4,
+        out_indices=(3,),
+        frozen_stages=1,
+        norm_cfg=dict(type='BN', requires_grad=False),
+        norm_eval=True,
+        style='pytorch'),
+    img_neck=dict(
+        type='FPN',
+        in_channels=[2048],
+        out_channels=_dim_,
+        start_level=0,
+        add_extra_convs='on_output',
+        num_outs=_num_levels_,
+        relu_before_extra_convs=True),
+    pts_bbox_head=dict(
+        type='MapTRv2Head',
+        bev_h=bev_h_,
+        bev_w=bev_w_,
+        num_query=900,
+        num_vec_one2one=50,
+        num_vec_one2many=300,
+        k_one2many=6,
+        num_pts_per_vec=fixed_ptsnum_per_pred_line, # one bbox
+        num_pts_per_gt_vec=fixed_ptsnum_per_gt_line,
+        dir_interval=1,
+        query_embed_type='instance_pts',
+        transform_method='minmax',
+        gt_shift_pts_pattern='v2',
+        num_classes=num_map_classes,
+        in_channels=_dim_,
+        sync_cls_avg_factor=True,
+        with_box_refine=True,
+        as_two_stage=False,
+        code_size=2,
+        code_weights=[1.0, 1.0, 1.0, 1.0],
+        aux_seg=aux_seg_cfg,
+        # z_cfg=z_cfg,
+        transformer=dict(
+            type='MapTRPerceptionTransformer',
+            rotate_prev_bev=True,
+            use_shift=True,
+            use_can_bus=True,
+            embed_dims=_dim_,
+            encoder=dict(
+                type='LSSTransform',
+                in_channels=_dim_,
+                out_channels=_dim_,
+                feat_down_sample=32,
+                pc_range=point_cloud_range,
+                voxel_size=voxel_size,
+                dbound=dbound,
+                downsample=2,
+                loss_depth_weight=3.0,
+                depthnet_cfg=dict(use_dcn=False, with_cp=False, aspp_mid_channels=96),
+                grid_config=grid_config,),
+            decoder=dict(
+                type='MapTRDecoder',
+                num_layers=6,
+                return_intermediate=True,
+                transformerlayers=dict(
+                    type='DecoupledDetrTransformerDecoderLayer',
+                    num_vec=num_vec,
+                    num_pts_per_vec=fixed_ptsnum_per_pred_line,
+                    attn_cfgs=[
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                        dict(
+                            type='MultiheadAttention',
+                            embed_dims=_dim_,
+                            num_heads=8,
+                            dropout=0.1),
+                         dict(
+                            type='CustomMSDeformableAttention',
+                            embed_dims=_dim_,
+                            num_levels=1),
+                    ],
+
+                    feedforward_channels=_ffn_dim_,
+                    ffn_dropout=0.1,
+                    operation_order=('self_attn', 'norm', 'self_attn', 'norm','cross_attn', 'norm',
+                                     'ffn', 'norm')))),
+        bbox_coder=dict(
+            type='MapTRNMSFreeCoder',
+            # post_center_range=[-61.2, -61.2, -10.0, 61.2, 61.2, 10.0],
+            post_center_range=[-20, -35, -20, -35, 20, 35, 20, 35],
+            pc_range=point_cloud_range,
+            max_num=50,
+            voxel_size=voxel_size,
+            num_classes=num_map_classes),
+        positional_encoding=dict(
+            type='LearnedPositionalEncoding',
+            num_feats=_pos_dim_,
+            row_num_embed=bev_h_,
+            col_num_embed=bev_w_,
+            ),
+        loss_cls=dict(
+            type='FocalLoss',
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            loss_weight=2.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=0.0),
+        loss_iou=dict(type='GIoULoss', loss_weight=0.0),
+        loss_pts=dict(type='PtsL1Loss', 
+                      loss_weight=5.0),
+        loss_dir=dict(type='PtsDirCosLoss', loss_weight=0.005),
+        loss_seg=dict(type='SimpleLoss', 
+            pos_weight=4.0,
+            loss_weight=1.0),
+        loss_pv_seg=dict(type='SimpleLoss', 
+                    pos_weight=1.0,
+                    loss_weight=2.0),),
+    # model training and testing settings
+    train_cfg=dict(pts=dict(
+        grid_size=[512, 512, 1],
+        voxel_size=voxel_size,
+        point_cloud_range=point_cloud_range,
+        out_size_factor=4,
+        assigner=dict(
+            type='MapTRAssigner',
+            cls_cost=dict(type='FocalLossCost', weight=2.0),
+            reg_cost=dict(type='BBoxL1Cost', weight=0.0, box_format='xywh'),
+            # reg_cost=dict(type='BBox3DL1Cost', weight=0.25),
+            # iou_cost=dict(type='IoUCost', weight=1.0), # Fake cost. This is just to make it compatible with DETR head.
+            iou_cost=dict(type='IoUCost', iou_mode='giou', weight=0.0),
+            pts_cost=dict(type='OrderedPtsL1Cost', 
+                      weight=5),
+            pc_range=point_cloud_range))))
+
+dataset_type = 'CustomNuScenesOfflineLocalMapDataset'
+data_root = 'data/nuscenes/'
+file_client_args = dict(backend='disk')
+
+
+train_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='PhotoMetricDistortionMultiViewImage'),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        file_client_args=file_client_args),
+    dict(type='CustomPointToMultiViewDepth', downsample=1, grid_config=grid_config),
+    dict(type='PadMultiViewImageDepth', size_divisor=32), 
+    dict(type='DefaultFormatBundle3D', with_gt=False, with_label=False,class_names=map_classes),
+    dict(type='CustomCollect3D', keys=['img', 'gt_depth'])
+]
+
+test_pipeline = [
+    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(type='RandomScaleImageMultiViewImage', scales=[0.5]),
+    dict(type='NormalizeMultiviewImage', **img_norm_cfg),
+   
+    dict(
+        type='MultiScaleFlipAug3D',
+        img_scale=(1600, 900),
+        pts_scale_ratio=1,
+        flip=False,
+        transforms=[
+            dict(type='PadMultiViewImage', size_divisor=32),
+            dict(
+                type='DefaultFormatBundle3D', 
+                with_gt=False, 
+                with_label=False,
+                class_names=map_classes),
+            dict(type='CustomCollect3D', keys=['img'])
+        ])
+]
+
+data = dict(
+    samples_per_gpu=8,
+    workers_per_gpu=32,
+    # 最优
+    #samples_per_gpu=12,
+    #workers_per_gpu=48, # TODO
+    train=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_map_infos_temporal_train.pkl',
+        pipeline=train_pipeline,
+        classes=class_names,
+        modality=input_modality,
+        aux_seg=aux_seg_cfg,
+        test_mode=False,
+        use_valid_flag=True,
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        queue_length=queue_length,
+        # we use box_type_3d='LiDAR' in kitti and nuscenes dataset
+        # and box_type_3d='Depth' in sunrgbd and scannet dataset.
+        box_type_3d='LiDAR'),
+    val=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_map_infos_temporal_val.pkl',
+        map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+        pipeline=test_pipeline,  bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        classes=class_names, modality=input_modality, samples_per_gpu=1),
+    test=dict(
+        type=dataset_type,
+        data_root=data_root,
+        ann_file=data_root + 'nuscenes_map_infos_temporal_val.pkl',
+        map_ann_file=data_root + 'nuscenes_map_anns_val.json',
+        pipeline=test_pipeline, 
+        bev_size=(bev_h_, bev_w_),
+        pc_range=point_cloud_range,
+        fixed_ptsnum_per_line=fixed_ptsnum_per_gt_line,
+        eval_use_same_gt_sample_num_flag=eval_use_same_gt_sample_num_flag,
+        padding_value=-10000,
+        map_classes=map_classes,
+        classes=class_names, 
+        modality=input_modality),
+    shuffler_sampler=dict(type='DistributedGroupSampler'),
+    nonshuffler_sampler=dict(type='DistributedSampler')
+)
+
+optimizer = dict(
+    type='AdamW',
+    #type='Miopen_AdamW',
+    lr=6e-4,
+    paramwise_cfg=dict(
+        custom_keys={
+            'img_backbone': dict(lr_mult=0.1),
+        }),
+    weight_decay=0.01)
+
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='CosineAnnealing',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=1.0 / 3,
+    min_lr_ratio=1e-3)
+total_epochs = 1   #24
+evaluation = dict(interval=2, pipeline=test_pipeline, metric='chamfer',
+                  save_best='NuscMap_chamfer/mAP', rule='greater')
+# total_epochs = 50
+# evaluation = dict(interval=1, pipeline=test_pipeline)
+
+runner = dict(type='EpochBasedRunner', max_epochs=total_epochs)
+
+log_config = dict(
+    interval=1,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        dict(type='TensorboardLoggerHook')
+    ])
+fp16 = dict(loss_scale=512.)
+checkpoint_config = dict(max_keep_ckpts=1, interval=2)
+find_unused_parameters=True