update codes

mmdet/configs/maskformer/maskformer_r50_ms_16xb1_75e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmcv.transforms import RandomChoice, RandomChoiceResize
+from mmengine.config import read_base
+from mmengine.model.weight_init import PretrainedInit
+from mmengine.optim.optimizer import OptimWrapper
+from mmengine.optim.scheduler import MultiStepLR
+from mmengine.runner import EpochBasedTrainLoop, TestLoop, ValLoop
+from torch.nn.modules.activation import ReLU
+from torch.nn.modules.batchnorm import BatchNorm2d
+from torch.nn.modules.normalization import GroupNorm
+from torch.optim.adamw import AdamW
+
+from mmdet.datasets.transforms.transforms import RandomCrop
+from mmdet.models import MaskFormer
+from mmdet.models.backbones import ResNet
+from mmdet.models.data_preprocessors.data_preprocessor import \
+    DetDataPreprocessor
+from mmdet.models.dense_heads.maskformer_head import MaskFormerHead
+from mmdet.models.layers.pixel_decoder import TransformerEncoderPixelDecoder
+from mmdet.models.losses import CrossEntropyLoss, DiceLoss, FocalLoss
+from mmdet.models.seg_heads.panoptic_fusion_heads import MaskFormerFusionHead
+from mmdet.models.task_modules.assigners.hungarian_assigner import \
+    HungarianAssigner
+from mmdet.models.task_modules.assigners.match_cost import (ClassificationCost,
+                                                            DiceCost,
+                                                            FocalLossCost)
+from mmdet.models.task_modules.samplers import MaskPseudoSampler
+
+with read_base():
+    from .._base_.datasets.coco_panoptic import *
+    from .._base_.default_runtime import *
+
+data_preprocessor = dict(
+    type=DetDataPreprocessor,
+    mean=[123.675, 116.28, 103.53],
+    std=[58.395, 57.12, 57.375],
+    bgr_to_rgb=True,
+    pad_size_divisor=1,
+    pad_mask=True,
+    mask_pad_value=0,
+    pad_seg=True,
+    seg_pad_value=255)
+
+num_things_classes = 80
+num_stuff_classes = 53
+num_classes = num_things_classes + num_stuff_classes
+model = dict(
+    type=MaskFormer,
+    data_preprocessor=data_preprocessor,
+    backbone=dict(
+        type=ResNet,
+        depth=50,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=-1,
+        norm_cfg=dict(type=BatchNorm2d, requires_grad=False),
+        norm_eval=True,
+        style='pytorch',
+        init_cfg=dict(
+            type=PretrainedInit, checkpoint='torchvision://resnet50')),
+    panoptic_head=dict(
+        type=MaskFormerHead,
+        in_channels=[256, 512, 1024, 2048],  # pass to pixel_decoder inside
+        feat_channels=256,
+        out_channels=256,
+        num_things_classes=num_things_classes,
+        num_stuff_classes=num_stuff_classes,
+        num_queries=100,
+        pixel_decoder=dict(
+            type=TransformerEncoderPixelDecoder,
+            norm_cfg=dict(type=GroupNorm, num_groups=32),
+            act_cfg=dict(type=ReLU),
+            encoder=dict(  # DetrTransformerEncoder
+                num_layers=6,
+                layer_cfg=dict(  # DetrTransformerEncoderLayer
+                    self_attn_cfg=dict(  # MultiheadAttention
+                        embed_dims=256,
+                        num_heads=8,
+                        dropout=0.1,
+                        batch_first=True),
+                    ffn_cfg=dict(
+                        embed_dims=256,
+                        feedforward_channels=2048,
+                        num_fcs=2,
+                        ffn_drop=0.1,
+                        act_cfg=dict(type=ReLU, inplace=True)))),
+            positional_encoding=dict(num_feats=128, normalize=True)),
+        enforce_decoder_input_project=False,
+        positional_encoding=dict(num_feats=128, normalize=True),
+        transformer_decoder=dict(  # DetrTransformerDecoder
+            num_layers=6,
+            layer_cfg=dict(  # DetrTransformerDecoderLayer
+                self_attn_cfg=dict(  # MultiheadAttention
+                    embed_dims=256,
+                    num_heads=8,
+                    dropout=0.1,
+                    batch_first=True),
+                cross_attn_cfg=dict(  # MultiheadAttention
+                    embed_dims=256,
+                    num_heads=8,
+                    dropout=0.1,
+                    batch_first=True),
+                ffn_cfg=dict(
+                    embed_dims=256,
+                    feedforward_channels=2048,
+                    num_fcs=2,
+                    ffn_drop=0.1,
+                    act_cfg=dict(type=ReLU, inplace=True))),
+            return_intermediate=True),
+        loss_cls=dict(
+            type=CrossEntropyLoss,
+            use_sigmoid=False,
+            loss_weight=1.0,
+            reduction='mean',
+            class_weight=[1.0] * num_classes + [0.1]),
+        loss_mask=dict(
+            type=FocalLoss,
+            use_sigmoid=True,
+            gamma=2.0,
+            alpha=0.25,
+            reduction='mean',
+            loss_weight=20.0),
+        loss_dice=dict(
+            type=DiceLoss,
+            use_sigmoid=True,
+            activate=True,
+            reduction='mean',
+            naive_dice=True,
+            eps=1.0,
+            loss_weight=1.0)),
+    panoptic_fusion_head=dict(
+        type=MaskFormerFusionHead,
+        num_things_classes=num_things_classes,
+        num_stuff_classes=num_stuff_classes,
+        loss_panoptic=None,
+        init_cfg=None),
+    train_cfg=dict(
+        assigner=dict(
+            type=HungarianAssigner,
+            match_costs=[
+                dict(type=ClassificationCost, weight=1.0),
+                dict(type=FocalLossCost, weight=20.0, binary_input=True),
+                dict(type=DiceCost, weight=1.0, pred_act=True, eps=1.0)
+            ]),
+        sampler=dict(type=MaskPseudoSampler)),
+    test_cfg=dict(
+        panoptic_on=True,
+        # For now, the dataset does not support
+        # evaluating semantic segmentation metric.
+        semantic_on=False,
+        instance_on=False,
+        # max_per_image is for instance segmentation.
+        max_per_image=100,
+        object_mask_thr=0.8,
+        iou_thr=0.8,
+        # In MaskFormer's panoptic postprocessing,
+        # it will not filter masks whose score is smaller than 0.5 .
+        filter_low_score=False),
+    init_cfg=None)
+
+# dataset settings
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(
+        type=LoadPanopticAnnotations,
+        with_bbox=True,
+        with_mask=True,
+        with_seg=True),
+    dict(type=RandomFlip, prob=0.5),
+    # dict(type=Resize, scale=(1333, 800), keep_ratio=True),
+    dict(
+        type=RandomChoice,
+        transforms=[[
+            dict(
+                type=RandomChoiceResize,
+                scales=[(480, 1333), (512, 1333), (544, 1333), (576, 1333),
+                        (608, 1333), (640, 1333), (672, 1333), (704, 1333),
+                        (736, 1333), (768, 1333), (800, 1333)],
+                resize_type=Resize,
+                keep_ratio=True)
+        ],
+                    [
+                        dict(
+                            type=RandomChoiceResize,
+                            scales=[(400, 1333), (500, 1333), (600, 1333)],
+                            resize_type=Resize,
+                            keep_ratio=True),
+                        dict(
+                            type=RandomCrop,
+                            crop_type='absolute_range',
+                            crop_size=(384, 600),
+                            allow_negative_crop=True),
+                        dict(
+                            type=RandomChoiceResize,
+                            scales=[(480, 1333), (512, 1333), (544, 1333),
+                                    (576, 1333), (608, 1333), (640, 1333),
+                                    (672, 1333), (704, 1333), (736, 1333),
+                                    (768, 1333), (800, 1333)],
+                            resize_type=Resize,
+                            keep_ratio=True)
+                    ]]),
+    dict(type=PackDetInputs)
+]
+
+train_dataloader.update(
+    dict(batch_size=1, num_workers=1, dataset=dict(pipeline=train_pipeline)))
+
+val_dataloader.update(dict(batch_size=1, num_workers=1))
+
+test_dataloader = val_dataloader
+
+# optimizer
+optim_wrapper = dict(
+    type=OptimWrapper,
+    optimizer=dict(
+        type=AdamW,
+        lr=0.0001,
+        weight_decay=0.0001,
+        eps=1e-8,
+        betas=(0.9, 0.999)),
+    paramwise_cfg=dict(
+        custom_keys={
+            'backbone': dict(lr_mult=0.1, decay_mult=1.0),
+            'query_embed': dict(lr_mult=1.0, decay_mult=0.0)
+        },
+        norm_decay_mult=0.0),
+    clip_grad=dict(max_norm=0.01, norm_type=2))
+
+max_epochs = 75
+
+# learning rate
+param_scheduler = dict(
+    type=MultiStepLR,
+    begin=0,
+    end=max_epochs,
+    by_epoch=True,
+    milestones=[50],
+    gamma=0.1)
+
+train_cfg = dict(
+    type=EpochBasedTrainLoop, max_epochs=max_epochs, val_interval=1)
+val_cfg = dict(type=ValLoop)
+test_cfg = dict(type=TestLoop)
+
+# Default setting for scaling LR automatically
+#   - `enable` means enable scaling LR automatically
+#       or not by default.
+#   - `base_batch_size` = (16 GPUs) x (1 samples per GPU).
+auto_scale_lr = dict(enable=False, base_batch_size=16)

mmdet/configs/maskformer/maskformer_swin_l_p4_w12_64xb1_ms_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+from mmengine.optim.scheduler import LinearLR
+
+from mmdet.models.backbones import SwinTransformer
+from mmdet.models.layers import PixelDecoder
+
+with read_base():
+    from .maskformer_r50_ms_16xb1_75e_coco import *
+
+pretrained = 'https://github.com/SwinTransformer/storage/releases/download/v1.0.0/swin_large_patch4_window12_384_22k.pth'  # noqa
+depths = [2, 2, 18, 2]
+model.update(
+    dict(
+        backbone=dict(
+            _delete_=True,
+            type=SwinTransformer,
+            pretrain_img_size=384,
+            embed_dims=192,
+            patch_size=4,
+            window_size=12,
+            mlp_ratio=4,
+            depths=depths,
+            num_heads=[6, 12, 24, 48],
+            qkv_bias=True,
+            qk_scale=None,
+            drop_rate=0.,
+            attn_drop_rate=0.,
+            drop_path_rate=0.3,
+            patch_norm=True,
+            out_indices=(0, 1, 2, 3),
+            with_cp=False,
+            convert_weights=True,
+            init_cfg=dict(type=PretrainedInit, checkpoint=pretrained)),
+        panoptic_head=dict(
+            in_channels=[192, 384, 768, 1536],  # pass to pixel_decoder inside
+            pixel_decoder=dict(
+                _delete_=True,
+                type=PixelDecoder,
+                norm_cfg=dict(type=GroupNorm, num_groups=32),
+                act_cfg=dict(type=ReLU)),
+            enforce_decoder_input_project=True)))
+
+# optimizer
+
+# weight_decay = 0.01
+# norm_weight_decay = 0.0
+# embed_weight_decay = 0.0
+embed_multi = dict(lr_mult=1.0, decay_mult=0.0)
+norm_multi = dict(lr_mult=1.0, decay_mult=0.0)
+custom_keys = {
+    'norm': norm_multi,
+    'absolute_pos_embed': embed_multi,
+    'relative_position_bias_table': embed_multi,
+    'query_embed': embed_multi
+}
+
+optim_wrapper.update(
+    dict(
+        optimizer=dict(lr=6e-5, weight_decay=0.01),
+        paramwise_cfg=dict(custom_keys=custom_keys, norm_decay_mult=0.0)))
+
+max_epochs = 300
+
+# learning rate
+param_scheduler = [
+    dict(type=LinearLR, start_factor=1e-6, by_epoch=False, begin=0, end=1500),
+    dict(
+        type=MultiStepLR,
+        begin=0,
+        end=max_epochs,
+        by_epoch=True,
+        milestones=[250],
+        gamma=0.1)
+]
+
+train_cfg.update(dict(max_epochs=max_epochs))
+
+# NOTE: `auto_scale_lr` is for automatically scaling LR,
+# USER SHOULD NOT CHANGE ITS VALUES.
+# base_batch_size = (64 GPUs) x (1 samples per GPU)
+auto_scale_lr.update(dict(base_batch_size=64))

mmdet/configs/panoptic_fpn/panoptic_fpn_r101_fpn_1x_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+from mmengine.model.weight_init import PretrainedInit
+
+with read_base():
+    from .panoptic_fpn_r50_fpn_1x_coco import *
+
+model.update(
+    dict(
+        backbone=dict(
+            depth=101,
+            init_cfg=dict(
+                type=PretrainedInit, checkpoint='torchvision://resnet101'))))

mmdet/configs/panoptic_fpn/panoptic_fpn_r101_fpn_ms_3x_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+from mmengine.model.weight_init import PretrainedInit
+
+with read_base():
+    from .panoptic_fpn_r50_fpn_ms_3x_coco import *
+
+model.update(
+    dict(
+        backbone=dict(
+            depth=101,
+            init_cfg=dict(
+                type=PretrainedInit, checkpoint='torchvision://resnet101'))))

mmdet/configs/panoptic_fpn/panoptic_fpn_r50_fpn_1x_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .._base_.models.mask_rcnn_r50_fpn import *
+    from .._base_.datasets.coco_panoptic import *
+    from .._base_.schedules.schedule_1x import *
+    from .._base_.default_runtime import *
+
+from mmcv.ops import nms
+from torch.nn import GroupNorm
+
+from mmdet.models.data_preprocessors.data_preprocessor import \
+    DetDataPreprocessor
+from mmdet.models.detectors.panoptic_fpn import PanopticFPN
+from mmdet.models.losses.cross_entropy_loss import CrossEntropyLoss
+from mmdet.models.seg_heads.panoptic_fpn_head import PanopticFPNHead
+from mmdet.models.seg_heads.panoptic_fusion_heads import HeuristicFusionHead
+
+model.update(
+    dict(
+        type=PanopticFPN,
+        data_preprocessor=dict(
+            type=DetDataPreprocessor,
+            mean=[123.675, 116.28, 103.53],
+            std=[58.395, 57.12, 57.375],
+            bgr_to_rgb=True,
+            pad_size_divisor=32,
+            pad_mask=True,
+            mask_pad_value=0,
+            pad_seg=True,
+            seg_pad_value=255),
+        semantic_head=dict(
+            type=PanopticFPNHead,
+            num_things_classes=80,
+            num_stuff_classes=53,
+            in_channels=256,
+            inner_channels=128,
+            start_level=0,
+            end_level=4,
+            norm_cfg=dict(type=GroupNorm, num_groups=32, requires_grad=True),
+            conv_cfg=None,
+            loss_seg=dict(
+                type=CrossEntropyLoss, ignore_index=255, loss_weight=0.5)),
+        panoptic_fusion_head=dict(
+            type=HeuristicFusionHead,
+            num_things_classes=80,
+            num_stuff_classes=53),
+        test_cfg=dict(
+            rcnn=dict(
+                score_thr=0.6,
+                nms=dict(type=nms, iou_threshold=0.5, class_agnostic=True),
+                max_per_img=100,
+                mask_thr_binary=0.5),
+            # used in HeuristicFusionHead
+            panoptic=dict(mask_overlap=0.5, stuff_area_limit=4096))))
+
+# Forced to remove NumClassCheckHook
+custom_hooks = []

mmdet/configs/panoptic_fpn/panoptic_fpn_r50_fpn_ms_3x_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+from mmengine.optim.scheduler.lr_scheduler import LinearLR, MultiStepLR
+
+with read_base():
+    from .panoptic_fpn_r50_fpn_1x_coco import *
+
+from mmcv.transforms import RandomResize
+from mmcv.transforms.loading import LoadImageFromFile
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import LoadPanopticAnnotations
+from mmdet.datasets.transforms.transforms import RandomFlip
+
+# In mstrain 3x config, img_scale=[(1333, 640), (1333, 800)],
+# multiscale_mode='range'
+train_pipeline = [
+    dict(type=LoadImageFromFile),
+    dict(
+        type=LoadPanopticAnnotations,
+        with_bbox=True,
+        with_mask=True,
+        with_seg=True),
+    dict(type=RandomResize, scale=[(1333, 640), (1333, 800)], keep_ratio=True),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=PackDetInputs)
+]
+
+train_dataloader.update(dict(dataset=dict(pipeline=train_pipeline)))
+
+# TODO: Use RepeatDataset to speed up training
+# training schedule for 3x
+train_cfg.update(dict(max_epochs=36, val_interval=3))
+
+# learning rate
+param_scheduler = [
+    dict(type=LinearLR, start_factor=0.001, by_epoch=False, begin=0, end=500),
+    dict(
+        type=MultiStepLR,
+        begin=0,
+        end=36,
+        by_epoch=True,
+        milestones=[24, 33],
+        gamma=0.1)
+]

mmdet/configs/qdtrack/qdtrack_faster_rcnn_r50_fpn_4e_base.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .._base_.models.faster_rcnn_r50_fpn import *
+    from .._base_.models.faster_rcnn_r50_fpn import model
+    from .._base_.default_runtime import *
+
+from mmcv.ops import RoIAlign
+from mmengine.hooks import LoggerHook, SyncBuffersHook
+from mmengine.model.weight_init import PretrainedInit
+from mmengine.optim import MultiStepLR, OptimWrapper
+from mmengine.runner.runner import EpochBasedTrainLoop, TestLoop, ValLoop
+from torch.nn.modules.batchnorm import BatchNorm2d
+from torch.nn.modules.normalization import GroupNorm
+from torch.optim import SGD
+
+from mmdet.engine.hooks import TrackVisualizationHook
+from mmdet.models import (QDTrack, QuasiDenseEmbedHead, QuasiDenseTracker,
+                          QuasiDenseTrackHead, SingleRoIExtractor,
+                          TrackDataPreprocessor)
+from mmdet.models.losses import (L1Loss, MarginL2Loss,
+                                 MultiPosCrossEntropyLoss, SmoothL1Loss)
+from mmdet.models.task_modules import (CombinedSampler,
+                                       InstanceBalancedPosSampler,
+                                       MaxIoUAssigner, RandomSampler)
+from mmdet.visualization import TrackLocalVisualizer
+
+detector = model
+detector.pop('data_preprocessor')
+
+detector['backbone'].update(
+    dict(
+        norm_cfg=dict(type=BatchNorm2d, requires_grad=False),
+        style='caffe',
+        init_cfg=dict(
+            type=PretrainedInit,
+            checkpoint='open-mmlab://detectron2/resnet50_caffe')))
+detector.rpn_head.loss_bbox.update(
+    dict(type=SmoothL1Loss, beta=1.0 / 9.0, loss_weight=1.0))
+detector.rpn_head.bbox_coder.update(dict(clip_border=False))
+detector.roi_head.bbox_head.update(dict(num_classes=1))
+detector.roi_head.bbox_head.bbox_coder.update(dict(clip_border=False))
+detector['init_cfg'] = dict(
+    type=PretrainedInit,
+    checkpoint=  # noqa: E251
+    'https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/'
+    'faster_rcnn_r50_fpn_1x_coco-person/'
+    'faster_rcnn_r50_fpn_1x_coco-person_20201216_175929-d022e227.pth'
+    # noqa: E501
+)
+del model
+
+model = dict(
+    type=QDTrack,
+    data_preprocessor=dict(
+        type=TrackDataPreprocessor,
+        mean=[103.530, 116.280, 123.675],
+        std=[1.0, 1.0, 1.0],
+        bgr_to_rgb=False,
+        pad_size_divisor=32),
+    detector=detector,
+    track_head=dict(
+        type=QuasiDenseTrackHead,
+        roi_extractor=dict(
+            type=SingleRoIExtractor,
+            roi_layer=dict(type=RoIAlign, output_size=7, sampling_ratio=0),
+            out_channels=256,
+            featmap_strides=[4, 8, 16, 32]),
+        embed_head=dict(
+            type=QuasiDenseEmbedHead,
+            num_convs=4,
+            num_fcs=1,
+            embed_channels=256,
+            norm_cfg=dict(type=GroupNorm, num_groups=32),
+            loss_track=dict(type=MultiPosCrossEntropyLoss, loss_weight=0.25),
+            loss_track_aux=dict(
+                type=MarginL2Loss,
+                neg_pos_ub=3,
+                pos_margin=0,
+                neg_margin=0.1,
+                hard_mining=True,
+                loss_weight=1.0)),
+        loss_bbox=dict(type=L1Loss, loss_weight=1.0),
+        train_cfg=dict(
+            assigner=dict(
+                type=MaxIoUAssigner,
+                pos_iou_thr=0.7,
+                neg_iou_thr=0.5,
+                min_pos_iou=0.5,
+                match_low_quality=False,
+                ignore_iof_thr=-1),
+            sampler=dict(
+                type=CombinedSampler,
+                num=256,
+                pos_fraction=0.5,
+                neg_pos_ub=3,
+                add_gt_as_proposals=True,
+                pos_sampler=dict(type=InstanceBalancedPosSampler),
+                neg_sampler=dict(type=RandomSampler)))),
+    tracker=dict(
+        type=QuasiDenseTracker,
+        init_score_thr=0.9,
+        obj_score_thr=0.5,
+        match_score_thr=0.5,
+        memo_tracklet_frames=30,
+        memo_backdrop_frames=1,
+        memo_momentum=0.8,
+        nms_conf_thr=0.5,
+        nms_backdrop_iou_thr=0.3,
+        nms_class_iou_thr=0.7,
+        with_cats=True,
+        match_metric='bisoftmax'))
+# optimizer
+optim_wrapper = dict(
+    type=OptimWrapper,
+    optimizer=dict(type=SGD, lr=0.02, momentum=0.9, weight_decay=0.0001),
+    clip_grad=dict(max_norm=35, norm_type=2))
+# learning policy
+param_scheduler = [
+    dict(type=MultiStepLR, begin=0, end=4, by_epoch=True, milestones=[3])
+]
+
+# runtime settings
+train_cfg = dict(type=EpochBasedTrainLoop, max_epochs=4, val_interval=4)
+val_cfg = dict(type=ValLoop)
+test_cfg = dict(type=TestLoop)
+
+default_hooks.update(
+    logger=dict(type=LoggerHook, interval=50),
+    visualization=dict(type=TrackVisualizationHook, draw=False))
+
+visualizer.update(
+    type=TrackLocalVisualizer, vis_backends=vis_backends, name='visualizer')
+
+# custom hooks
+custom_hooks = [
+    # Synchronize model buffers such as running_mean and running_var in BN
+    # at the end of each epoch
+    dict(type=SyncBuffersHook)
+]

mmdet/configs/qdtrack/qdtrack_faster_rcnn_r50_fpn_8xb2-4e_mot17halftrain_test-mot17halfval.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from .._base_.datasets.mot_challenge import *
+    from .qdtrack_faster_rcnn_r50_fpn_4e_base import *
+
+from mmdet.evaluation import CocoVideoMetric, MOTChallengeMetric
+
+# evaluator
+val_evaluator = [
+    dict(type=CocoVideoMetric, metric=['bbox'], classwise=True),
+    dict(type=MOTChallengeMetric, metric=['HOTA', 'CLEAR', 'Identity'])
+]

mmdet/configs/retinanet/retinanet_r50_fpn_1x_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .._base_.models.retinanet_r50_fpn import *
+    from .._base_.datasets.coco_detection import *
+    from .._base_.schedules.schedule_1x import *
+    from .._base_.default_runtime import *
+    from .retinanet_tta import *
+
+from torch.optim.sgd import SGD
+
+# optimizer
+optim_wrapper.update(
+    dict(optimizer=dict(type=SGD, lr=0.01, momentum=0.9, weight_decay=0.0001)))

mmdet/configs/retinanet/retinanet_tta.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import TestTimeAug
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import LoadAnnotations
+from mmdet.datasets.transforms.transforms import RandomFlip, Resize
+from mmdet.models.test_time_augs.det_tta import DetTTAModel
+
+tta_model = dict(
+    type=DetTTAModel,
+    tta_cfg=dict(nms=dict(type='nms', iou_threshold=0.5), max_per_img=100))
+
+img_scales = [(1333, 800), (666, 400), (2000, 1200)]
+tta_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=None),
+    dict(
+        type=TestTimeAug,
+        transforms=[
+            [dict(type=Resize, scale=s, keep_ratio=True) for s in img_scales],
+            [dict(type=RandomFlip, prob=1.),
+             dict(type=RandomFlip, prob=0.)],
+            [dict(type=LoadAnnotations, with_bbox=True)],
+            [
+                dict(
+                    type=PackDetInputs,
+                    meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                               'scale_factor', 'flip', 'flip_direction'))
+            ]
+        ])
+]

mmdet/configs/rtmdet/rtmdet_ins_l_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_l_8xb32_300e_coco import *
+
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import RandomResize
+from mmengine.hooks.ema_hook import EMAHook
+from torch.nn.modules.activation import SiLU
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import (FilterAnnotations,
+                                               LoadAnnotations)
+from mmdet.datasets.transforms.transforms import (CachedMixUp, CachedMosaic,
+                                                  Pad, RandomCrop, RandomFlip,
+                                                  Resize, YOLOXHSVRandomAug)
+from mmdet.engine.hooks.pipeline_switch_hook import PipelineSwitchHook
+from mmdet.models.dense_heads.rtmdet_ins_head import RTMDetInsSepBNHead
+from mmdet.models.layers.ema import ExpMomentumEMA
+from mmdet.models.losses.dice_loss import DiceLoss
+from mmdet.models.losses.gfocal_loss import QualityFocalLoss
+from mmdet.models.losses.iou_loss import GIoULoss
+from mmdet.models.task_modules.coders.distance_point_bbox_coder import \
+    DistancePointBBoxCoder
+from mmdet.models.task_modules.prior_generators.point_generator import \
+    MlvlPointGenerator
+
+model.merge(
+    dict(
+        bbox_head=dict(
+            _delete_=True,
+            type=RTMDetInsSepBNHead,
+            num_classes=80,
+            in_channels=256,
+            stacked_convs=2,
+            share_conv=True,
+            pred_kernel_size=1,
+            feat_channels=256,
+            act_cfg=dict(type=SiLU, inplace=True),
+            norm_cfg=dict(type='SyncBN', requires_grad=True),
+            anchor_generator=dict(
+                type=MlvlPointGenerator, offset=0, strides=[8, 16, 32]),
+            bbox_coder=dict(type=DistancePointBBoxCoder),
+            loss_cls=dict(
+                type=QualityFocalLoss,
+                use_sigmoid=True,
+                beta=2.0,
+                loss_weight=1.0),
+            loss_bbox=dict(type=GIoULoss, loss_weight=2.0),
+            loss_mask=dict(
+                type=DiceLoss, loss_weight=2.0, eps=5e-6, reduction='mean')),
+        test_cfg=dict(
+            nms_pre=1000,
+            min_bbox_size=0,
+            score_thr=0.05,
+            nms=dict(type='nms', iou_threshold=0.6),
+            max_per_img=100,
+            mask_thr_binary=0.5),
+    ))
+
+train_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(
+        type=LoadAnnotations, with_bbox=True, with_mask=True, poly2mask=False),
+    dict(type=CachedMosaic, img_scale=(640, 640), pad_val=114.0),
+    dict(
+        type=RandomResize,
+        scale=(1280, 1280),
+        ratio_range=(0.1, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(
+        type=RandomCrop,
+        crop_size=(640, 640),
+        recompute_bbox=True,
+        allow_negative_crop=True),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type=CachedMixUp,
+        img_scale=(640, 640),
+        ratio_range=(1.0, 1.0),
+        max_cached_images=20,
+        pad_val=(114, 114, 114)),
+    dict(type=FilterAnnotations, min_gt_bbox_wh=(1, 1)),
+    dict(type=PackDetInputs)
+]
+
+train_dataloader.update(
+    dict(pin_memory=True, dataset=dict(pipeline=train_pipeline)))
+
+train_pipeline_stage2 = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(
+        type=LoadAnnotations, with_bbox=True, with_mask=True, poly2mask=False),
+    dict(
+        type=RandomResize,
+        scale=(640, 640),
+        ratio_range=(0.1, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(
+        type=RandomCrop,
+        crop_size=(640, 640),
+        recompute_bbox=True,
+        allow_negative_crop=True),
+    dict(type=FilterAnnotations, min_gt_bbox_wh=(1, 1)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(type=PackDetInputs)
+]
+custom_hooks = [
+    dict(
+        type=EMAHook,
+        ema_type=ExpMomentumEMA,
+        momentum=0.0002,
+        update_buffers=True,
+        priority=49),
+    dict(
+        type=PipelineSwitchHook,
+        switch_epoch=280,
+        switch_pipeline=train_pipeline_stage2)
+]
+
+val_evaluator.update(dict(metric=['bbox', 'segm']))
+test_evaluator = val_evaluator

mmdet/configs/rtmdet/rtmdet_ins_m_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_ins_l_8xb32_300e_coco import *
+
+model.update(
+    dict(
+        backbone=dict(deepen_factor=0.67, widen_factor=0.75),
+        neck=dict(
+            in_channels=[192, 384, 768], out_channels=192, num_csp_blocks=2),
+        bbox_head=dict(in_channels=192, feat_channels=192)))

mmdet/configs/rtmdet/rtmdet_ins_s_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_ins_l_8xb32_300e_coco import *
+
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import RandomResize
+from mmengine.hooks.ema_hook import EMAHook
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import (FilterAnnotations,
+                                               LoadAnnotations)
+from mmdet.datasets.transforms.transforms import (CachedMixUp, CachedMosaic,
+                                                  Pad, RandomCrop, RandomFlip,
+                                                  Resize, YOLOXHSVRandomAug)
+from mmdet.engine.hooks.pipeline_switch_hook import PipelineSwitchHook
+from mmdet.models.layers.ema import ExpMomentumEMA
+
+checkpoint = 'https://download.openmmlab.com/mmdetection/v3.0/rtmdet/cspnext_rsb_pretrain/cspnext-s_imagenet_600e.pth'  # noqa
+model.update(
+    dict(
+        backbone=dict(
+            deepen_factor=0.33,
+            widen_factor=0.5,
+            init_cfg=dict(
+                type='Pretrained', prefix='backbone.', checkpoint=checkpoint)),
+        neck=dict(
+            in_channels=[128, 256, 512], out_channels=128, num_csp_blocks=1),
+        bbox_head=dict(in_channels=128, feat_channels=128)))
+
+train_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(
+        type=LoadAnnotations, with_bbox=True, with_mask=True, poly2mask=False),
+    dict(type=CachedMosaic, img_scale=(640, 640), pad_val=114.0),
+    dict(
+        type=RandomResize,
+        scale=(1280, 1280),
+        ratio_range=(0.5, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(
+        type=RandomCrop,
+        crop_size=(640, 640),
+        recompute_bbox=True,
+        allow_negative_crop=True),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type=CachedMixUp,
+        img_scale=(640, 640),
+        ratio_range=(1.0, 1.0),
+        max_cached_images=20,
+        pad_val=(114, 114, 114)),
+    dict(type=FilterAnnotations, min_gt_bbox_wh=(1, 1)),
+    dict(type=PackDetInputs)
+]
+
+train_pipeline_stage2 = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(
+        type=LoadAnnotations, with_bbox=True, with_mask=True, poly2mask=False),
+    dict(
+        type=RandomResize,
+        scale=(640, 640),
+        ratio_range=(0.5, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(
+        type=RandomCrop,
+        crop_size=(640, 640),
+        recompute_bbox=True,
+        allow_negative_crop=True),
+    dict(type=FilterAnnotations, min_gt_bbox_wh=(1, 1)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(type=PackDetInputs)
+]
+
+train_dataloader.update(dict(dataset=dict(pipeline=train_pipeline)))
+
+custom_hooks = [
+    dict(
+        type=EMAHook,
+        ema_type=ExpMomentumEMA,
+        momentum=0.0002,
+        update_buffers=True,
+        priority=49),
+    dict(
+        type=PipelineSwitchHook,
+        switch_epoch=280,
+        switch_pipeline=train_pipeline_stage2)
+]

mmdet/configs/rtmdet/rtmdet_ins_tiny_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_ins_s_8xb32_300e_coco import *
+
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import RandomResize
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import (FilterAnnotations,
+                                               LoadAnnotations)
+from mmdet.datasets.transforms.transforms import (CachedMixUp, CachedMosaic,
+                                                  Pad, RandomCrop, RandomFlip,
+                                                  Resize, YOLOXHSVRandomAug)
+
+checkpoint = 'https://download.openmmlab.com/mmdetection/v3.0/rtmdet/cspnext_rsb_pretrain/cspnext-tiny_imagenet_600e.pth'  # noqa
+
+model.update(
+    dict(
+        backbone=dict(
+            deepen_factor=0.167,
+            widen_factor=0.375,
+            init_cfg=dict(
+                type='Pretrained', prefix='backbone.', checkpoint=checkpoint)),
+        neck=dict(
+            in_channels=[96, 192, 384], out_channels=96, num_csp_blocks=1),
+        bbox_head=dict(in_channels=96, feat_channels=96)))
+
+train_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(
+        type=LoadAnnotations, with_bbox=True, with_mask=True, poly2mask=False),
+    dict(
+        type=CachedMosaic,
+        img_scale=(640, 640),
+        pad_val=114.0,
+        max_cached_images=20,
+        random_pop=False),
+    dict(
+        type=RandomResize,
+        scale=(1280, 1280),
+        ratio_range=(0.5, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=(640, 640)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type=CachedMixUp,
+        img_scale=(640, 640),
+        ratio_range=(1.0, 1.0),
+        max_cached_images=10,
+        random_pop=False,
+        pad_val=(114, 114, 114),
+        prob=0.5),
+    dict(type=FilterAnnotations, min_gt_bbox_wh=(1, 1)),
+    dict(type=PackDetInputs)
+]
+
+train_dataloader.update(dict(dataset=dict(pipeline=train_pipeline)))

mmdet/configs/rtmdet/rtmdet_ins_x_8xb16_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_ins_l_8xb32_300e_coco import *
+from mmengine.optim.scheduler.lr_scheduler import CosineAnnealingLR, LinearLR
+
+model.update(
+    dict(
+        backbone=dict(deepen_factor=1.33, widen_factor=1.25),
+        neck=dict(
+            in_channels=[320, 640, 1280], out_channels=320, num_csp_blocks=4),
+        bbox_head=dict(in_channels=320, feat_channels=320)))
+
+base_lr = 0.002
+
+# optimizer
+optim_wrapper.update(dict(optimizer=dict(lr=base_lr)))
+
+# learning rate
+param_scheduler = [
+    dict(
+        type=LinearLR, start_factor=1.0e-5, by_epoch=False, begin=0, end=1000),
+    dict(
+        # use cosine lr from 150 to 300 epoch
+        type=CosineAnnealingLR,
+        eta_min=base_lr * 0.05,
+        begin=max_epochs // 2,
+        end=max_epochs,
+        T_max=max_epochs // 2,
+        by_epoch=True,
+        convert_to_iter_based=True),
+]

mmdet/configs/rtmdet/rtmdet_l_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .._base_.default_runtime import *
+    from .._base_.schedules.schedule_1x import *
+    from .._base_.datasets.coco_detection import *
+    from .rtmdet_tta import *
+
+from mmcv.ops import nms
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import RandomResize
+from mmengine.hooks.ema_hook import EMAHook
+from mmengine.optim.optimizer.optimizer_wrapper import OptimWrapper
+from mmengine.optim.scheduler.lr_scheduler import CosineAnnealingLR, LinearLR
+from torch.nn import SyncBatchNorm
+from torch.nn.modules.activation import SiLU
+from torch.optim.adamw import AdamW
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import LoadAnnotations
+from mmdet.datasets.transforms.transforms import (CachedMixUp, CachedMosaic,
+                                                  Pad, RandomCrop, RandomFlip,
+                                                  Resize, YOLOXHSVRandomAug)
+from mmdet.engine.hooks.pipeline_switch_hook import PipelineSwitchHook
+from mmdet.models.backbones.cspnext import CSPNeXt
+from mmdet.models.data_preprocessors.data_preprocessor import \
+    DetDataPreprocessor
+from mmdet.models.dense_heads.rtmdet_head import RTMDetSepBNHead
+from mmdet.models.detectors.rtmdet import RTMDet
+from mmdet.models.layers.ema import ExpMomentumEMA
+from mmdet.models.losses.gfocal_loss import QualityFocalLoss
+from mmdet.models.losses.iou_loss import GIoULoss
+from mmdet.models.necks.cspnext_pafpn import CSPNeXtPAFPN
+from mmdet.models.task_modules.assigners.dynamic_soft_label_assigner import \
+    DynamicSoftLabelAssigner
+from mmdet.models.task_modules.coders.distance_point_bbox_coder import \
+    DistancePointBBoxCoder
+from mmdet.models.task_modules.prior_generators.point_generator import \
+    MlvlPointGenerator
+
+model = dict(
+    type=RTMDet,
+    data_preprocessor=dict(
+        type=DetDataPreprocessor,
+        mean=[103.53, 116.28, 123.675],
+        std=[57.375, 57.12, 58.395],
+        bgr_to_rgb=False,
+        batch_augments=None),
+    backbone=dict(
+        type=CSPNeXt,
+        arch='P5',
+        expand_ratio=0.5,
+        deepen_factor=1,
+        widen_factor=1,
+        channel_attention=True,
+        norm_cfg=dict(type=SyncBatchNorm),
+        act_cfg=dict(type=SiLU, inplace=True)),
+    neck=dict(
+        type=CSPNeXtPAFPN,
+        in_channels=[256, 512, 1024],
+        out_channels=256,
+        num_csp_blocks=3,
+        expand_ratio=0.5,
+        norm_cfg=dict(type=SyncBatchNorm),
+        act_cfg=dict(type=SiLU, inplace=True)),
+    bbox_head=dict(
+        type=RTMDetSepBNHead,
+        num_classes=80,
+        in_channels=256,
+        stacked_convs=2,
+        feat_channels=256,
+        anchor_generator=dict(
+            type=MlvlPointGenerator, offset=0, strides=[8, 16, 32]),
+        bbox_coder=dict(type=DistancePointBBoxCoder),
+        loss_cls=dict(
+            type=QualityFocalLoss, use_sigmoid=True, beta=2.0,
+            loss_weight=1.0),
+        loss_bbox=dict(type=GIoULoss, loss_weight=2.0),
+        with_objectness=False,
+        exp_on_reg=True,
+        share_conv=True,
+        pred_kernel_size=1,
+        norm_cfg=dict(type=SyncBatchNorm),
+        act_cfg=dict(type=SiLU, inplace=True)),
+    train_cfg=dict(
+        assigner=dict(type=DynamicSoftLabelAssigner, topk=13),
+        allowed_border=-1,
+        pos_weight=-1,
+        debug=False),
+    test_cfg=dict(
+        nms_pre=30000,
+        min_bbox_size=0,
+        score_thr=0.001,
+        nms=dict(type=nms, iou_threshold=0.65),
+        max_per_img=300),
+)
+
+train_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(type=LoadAnnotations, with_bbox=True),
+    dict(type=CachedMosaic, img_scale=(640, 640), pad_val=114.0),
+    dict(
+        type=RandomResize,
+        scale=(1280, 1280),
+        ratio_range=(0.1, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=(640, 640)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type=CachedMixUp,
+        img_scale=(640, 640),
+        ratio_range=(1.0, 1.0),
+        max_cached_images=20,
+        pad_val=(114, 114, 114)),
+    dict(type=PackDetInputs)
+]
+
+train_pipeline_stage2 = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(type=LoadAnnotations, with_bbox=True),
+    dict(
+        type=RandomResize,
+        scale=(640, 640),
+        ratio_range=(0.1, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=(640, 640)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(type=PackDetInputs)
+]
+
+test_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(type=Resize, scale=(640, 640), keep_ratio=True),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(type=LoadAnnotations, with_bbox=True),
+    dict(
+        type=PackDetInputs,
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor'))
+]
+
+train_dataloader.update(
+    dict(
+        batch_size=32,
+        num_workers=10,
+        batch_sampler=None,
+        pin_memory=True,
+        dataset=dict(pipeline=train_pipeline)))
+val_dataloader.update(
+    dict(batch_size=5, num_workers=10, dataset=dict(pipeline=test_pipeline)))
+test_dataloader = val_dataloader
+
+max_epochs = 300
+stage2_num_epochs = 20
+base_lr = 0.004
+interval = 10
+
+train_cfg.update(
+    dict(
+        max_epochs=max_epochs,
+        val_interval=interval,
+        dynamic_intervals=[(max_epochs - stage2_num_epochs, 1)]))
+
+val_evaluator.update(dict(proposal_nums=(100, 1, 10)))
+test_evaluator = val_evaluator
+
+# optimizer
+optim_wrapper = dict(
+    type=OptimWrapper,
+    optimizer=dict(type=AdamW, lr=base_lr, weight_decay=0.05),
+    paramwise_cfg=dict(
+        norm_decay_mult=0, bias_decay_mult=0, bypass_duplicate=True))
+
+# learning rate
+param_scheduler = [
+    dict(
+        type=LinearLR, start_factor=1.0e-5, by_epoch=False, begin=0, end=1000),
+    dict(
+        # use cosine lr from 150 to 300 epoch
+        type=CosineAnnealingLR,
+        eta_min=base_lr * 0.05,
+        begin=max_epochs // 2,
+        end=max_epochs,
+        T_max=max_epochs // 2,
+        by_epoch=True,
+        convert_to_iter_based=True),
+]
+
+# hooks
+default_hooks.update(
+    dict(
+        checkpoint=dict(
+            interval=interval,
+            max_keep_ckpts=3  # only keep latest 3 checkpoints
+        )))
+
+custom_hooks = [
+    dict(
+        type=EMAHook,
+        ema_type=ExpMomentumEMA,
+        momentum=0.0002,
+        update_buffers=True,
+        priority=49),
+    dict(
+        type=PipelineSwitchHook,
+        switch_epoch=max_epochs - stage2_num_epochs,
+        switch_pipeline=train_pipeline_stage2)
+]

mmdet/configs/rtmdet/rtmdet_m_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_l_8xb32_300e_coco import *
+
+model.update(
+    dict(
+        backbone=dict(deepen_factor=0.67, widen_factor=0.75),
+        neck=dict(
+            in_channels=[192, 384, 768], out_channels=192, num_csp_blocks=2),
+        bbox_head=dict(in_channels=192, feat_channels=192)))

mmdet/configs/rtmdet/rtmdet_s_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_l_8xb32_300e_coco import *
+
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import RandomResize
+from mmengine.hooks.ema_hook import EMAHook
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import LoadAnnotations
+from mmdet.datasets.transforms.transforms import (CachedMixUp, CachedMosaic,
+                                                  Pad, RandomCrop, RandomFlip,
+                                                  Resize, YOLOXHSVRandomAug)
+from mmdet.engine.hooks.pipeline_switch_hook import PipelineSwitchHook
+from mmdet.models.layers.ema import ExpMomentumEMA
+
+checkpoint = 'https://download.openmmlab.com/mmdetection/v3.0/rtmdet/cspnext_rsb_pretrain/cspnext-s_imagenet_600e.pth'  # noqa
+model.update(
+    dict(
+        backbone=dict(
+            deepen_factor=0.33,
+            widen_factor=0.5,
+            init_cfg=dict(
+                type='Pretrained', prefix='backbone.', checkpoint=checkpoint)),
+        neck=dict(
+            in_channels=[128, 256, 512], out_channels=128, num_csp_blocks=1),
+        bbox_head=dict(in_channels=128, feat_channels=128, exp_on_reg=False)))
+
+train_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(type=LoadAnnotations, with_bbox=True),
+    dict(type=CachedMosaic, img_scale=(640, 640), pad_val=114.0),
+    dict(
+        type=RandomResize,
+        scale=(1280, 1280),
+        ratio_range=(0.5, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=(640, 640)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type=CachedMixUp,
+        img_scale=(640, 640),
+        ratio_range=(1.0, 1.0),
+        max_cached_images=20,
+        pad_val=(114, 114, 114)),
+    dict(type=PackDetInputs)
+]
+
+train_pipeline_stage2 = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(type=LoadAnnotations, with_bbox=True),
+    dict(
+        type=RandomResize,
+        scale=(640, 640),
+        ratio_range=(0.5, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=(640, 640)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(type=PackDetInputs)
+]
+
+train_dataloader.update(dict(dataset=dict(pipeline=train_pipeline)))
+
+custom_hooks = [
+    dict(
+        type=EMAHook,
+        ema_type=ExpMomentumEMA,
+        momentum=0.0002,
+        update_buffers=True,
+        priority=49),
+    dict(
+        type=PipelineSwitchHook,
+        switch_epoch=280,
+        switch_pipeline=train_pipeline_stage2)
+]

mmdet/configs/rtmdet/rtmdet_tiny_8xb32_300e_coco.py

+# Copyright (c) OpenMMLab. All rights reserved.
+
+# Please refer to https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#a-pure-python-style-configuration-file-beta for more details. # noqa
+# mmcv >= 2.0.1
+# mmengine >= 0.8.0
+
+from mmengine.config import read_base
+
+with read_base():
+    from .rtmdet_s_8xb32_300e_coco import *
+
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import RandomResize
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import LoadAnnotations
+from mmdet.datasets.transforms.transforms import (CachedMixUp, CachedMosaic,
+                                                  Pad, RandomCrop, RandomFlip,
+                                                  Resize, YOLOXHSVRandomAug)
+
+checkpoint = 'https://download.openmmlab.com/mmdetection/v3.0/rtmdet/cspnext_rsb_pretrain/cspnext-tiny_imagenet_600e.pth'  # noqa
+
+model.update(
+    dict(
+        backbone=dict(
+            deepen_factor=0.167,
+            widen_factor=0.375,
+            init_cfg=dict(
+                type='Pretrained', prefix='backbone.', checkpoint=checkpoint)),
+        neck=dict(
+            in_channels=[96, 192, 384], out_channels=96, num_csp_blocks=1),
+        bbox_head=dict(in_channels=96, feat_channels=96, exp_on_reg=False)))
+
+train_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=backend_args),
+    dict(type=LoadAnnotations, with_bbox=True),
+    dict(
+        type=CachedMosaic,
+        img_scale=(640, 640),
+        pad_val=114.0,
+        max_cached_images=20,
+        random_pop=False),
+    dict(
+        type=RandomResize,
+        scale=(1280, 1280),
+        ratio_range=(0.5, 2.0),
+        resize_type=Resize,
+        keep_ratio=True),
+    dict(type=RandomCrop, crop_size=(640, 640)),
+    dict(type=YOLOXHSVRandomAug),
+    dict(type=RandomFlip, prob=0.5),
+    dict(type=Pad, size=(640, 640), pad_val=dict(img=(114, 114, 114))),
+    dict(
+        type=CachedMixUp,
+        img_scale=(640, 640),
+        ratio_range=(1.0, 1.0),
+        max_cached_images=10,
+        random_pop=False,
+        pad_val=(114, 114, 114),
+        prob=0.5),
+    dict(type=PackDetInputs)
+]
+
+train_dataloader.update(dict(dataset=dict(pipeline=train_pipeline)))

mmdet/configs/rtmdet/rtmdet_tta.py

+# Copyright (c) OpenMMLab. All rights reserved.
+from mmcv.transforms.loading import LoadImageFromFile
+from mmcv.transforms.processing import TestTimeAug
+
+from mmdet.datasets.transforms.formatting import PackDetInputs
+from mmdet.datasets.transforms.loading import LoadAnnotations
+from mmdet.datasets.transforms.transforms import Pad, RandomFlip, Resize
+from mmdet.models.test_time_augs.det_tta import DetTTAModel
+
+tta_model = dict(
+    type=DetTTAModel,
+    tta_cfg=dict(nms=dict(type='nms', iou_threshold=0.6), max_per_img=100))
+
+img_scales = [(640, 640), (320, 320), (960, 960)]
+
+tta_pipeline = [
+    dict(type=LoadImageFromFile, backend_args=None),
+    dict(
+        type=TestTimeAug,
+        transforms=[
+            [dict(type=Resize, scale=s, keep_ratio=True) for s in img_scales],
+            [
+                # ``RandomFlip`` must be placed before ``Pad``, otherwise
+                # bounding box coordinates after flipping cannot be
+                # recovered correctly.
+                dict(type=RandomFlip, prob=1.),
+                dict(type=RandomFlip, prob=0.)
+            ],
+            [
+                dict(
+                    type=Pad,
+                    size=(960, 960),
+                    pad_val=dict(img=(114, 114, 114))),
+            ],
+            [dict(type=LoadAnnotations, with_bbox=True)],
+            [
+                dict(
+                    type=PackDetInputs,
+                    meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                               'scale_factor', 'flip', 'flip_direction'))
+            ]
+        ])
+]