[FIX] replace DefaultFormatBundle/3D with Pack(3D)DetInputs (#1987)

* replace defaultformatbundle3d with pack3ddetinputs

* remove normalize, pad, imagetotensor from configs

* rm unused key 'img_norm_cfg'

* fix lint errors

* fix lint errors

* fix lint error

* xx

configs/_base_/datasets/lyft-3d-range100.py

@@ -39,8 +39,9 @@ train_pipeline = [
     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'])
+    dict(
+        type='Pack3DDetInputs',
+        keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
 ]
 test_pipeline = [
     dict(type='LoadPointsFromFile', coord_type='LIDAR', load_dim=5, use_dim=5),
@@ -59,23 +60,15 @@ test_pipeline = [
             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'])
-        ])
+        ]),
+    dict(type='Pack3DDetInputs', 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),
     dict(type='LoadPointsFromMultiSweeps', sweeps_num=10),
-    dict(
-        type='DefaultFormatBundle3D',
-        class_names=class_names,
-        with_label=False),
-    dict(type='Collect3D', keys=['points'])
+    dict(type='Pack3DDetInputs', keys=['points'])
 ]
 
 data = dict(

configs/_base_/datasets/nuim-instance.py

@@ -4,8 +4,6 @@ class_names = [
     'car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
     'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'
 ]
-img_norm_cfg = dict(
-    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
 
 file_client_args = dict(backend='disk')
 # Uncomment the following if use ceph or other file clients.
@@ -23,10 +21,7 @@ train_pipeline = [
         multiscale_mode='range',
         keep_ratio=True),
     dict(type='RandomFlip', flip_ratio=0.5),
-    dict(type='Normalize', **img_norm_cfg),
-    dict(type='Pad', size_divisor=32),
-    dict(type='DefaultFormatBundle'),
-    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
+    dict(type='PackDetInputs'),
 ]
 test_pipeline = [
     dict(type='LoadImageFromFile'),
@@ -37,11 +32,11 @@ test_pipeline = [
         transforms=[
             dict(type='Resize', keep_ratio=True),
             dict(type='RandomFlip'),
-            dict(type='Normalize', **img_norm_cfg),
-            dict(type='Pad', size_divisor=32),
-            dict(type='ImageToTensor', keys=['img']),
-            dict(type='Collect', keys=['img']),
-        ])
+        ]),
+    dict(
+        type='PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor')),
 ]
 data = dict(
     samples_per_gpu=2,

configs/centerpoint/README.md

@@ -91,12 +91,8 @@ test_pipeline = [
             dict(type='RandomFlip3D', sync_2d=False),
             dict(
                 type='PointsRangeFilter', point_cloud_range=point_cloud_range),
-            dict(
-                type='DefaultFormatBundle3D',
-                class_names=class_names,
-                with_label=False),
-            dict(type='Collect3D', keys=['points'])
-        ])
+        ]),
+    dict(type='Pack3DDetInputs', keys=['points'])
 ]
 
 data = dict(

configs/nuimages/htc_r50_fpn_1x_nuim.py

@@ -18,8 +18,6 @@ model = dict(
             loss_weight=0.2)))
 
 data_root = 'data/nuimages/'
-img_norm_cfg = dict(
-    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
 train_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(
@@ -30,13 +28,8 @@ train_pipeline = [
         multiscale_mode='range',
         keep_ratio=True),
     dict(type='RandomFlip', flip_ratio=0.5),
-    dict(type='Normalize', **img_norm_cfg),
-    dict(type='Pad', size_divisor=32),
     dict(type='SegRescale', scale_factor=1 / 8),
-    dict(type='DefaultFormatBundle'),
-    dict(
-        type='Collect',
-        keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks', 'gt_semantic_seg'])
+    dict(type='PackDetInputs')
 ]
 data = dict(
     train=dict(

configs/nuimages/mask-rcnn_r50_caffe_fpn_1x_nuim.py

@@ -8,9 +8,6 @@ model = dict(
     backbone=dict(norm_cfg=dict(requires_grad=False), style='caffe'),
     roi_head=dict(
         bbox_head=dict(num_classes=10), mask_head=dict(num_classes=10)))
-# use caffe img_norm
-img_norm_cfg = dict(
-    mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
 train_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
@@ -20,10 +17,7 @@ train_pipeline = [
         multiscale_mode='range',
         keep_ratio=True),
     dict(type='RandomFlip', flip_ratio=0.5),
-    dict(type='Normalize', **img_norm_cfg),
-    dict(type='Pad', size_divisor=32),
-    dict(type='DefaultFormatBundle'),
-    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
+    dict(type='PackDetInputs'),
 ]
 test_pipeline = [
     dict(type='LoadImageFromFile'),
@@ -34,11 +28,11 @@ test_pipeline = [
         transforms=[
             dict(type='Resize', keep_ratio=True),
             dict(type='RandomFlip'),
-            dict(type='Normalize', **img_norm_cfg),
-            dict(type='Pad', size_divisor=32),
-            dict(type='ImageToTensor', keys=['img']),
-            dict(type='Collect', keys=['img']),
-        ])
+        ]),
+    dict(
+        type='PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor')),
 ]
 data = dict(
     train=dict(pipeline=train_pipeline),

configs/nuimages/mask-rcnn_r50_caffe_fpn_coco-3x_1x_nuim.py

@@ -8,9 +8,6 @@ model = dict(
     backbone=dict(norm_cfg=dict(requires_grad=False), style='caffe'),
     roi_head=dict(
         bbox_head=dict(num_classes=10), mask_head=dict(num_classes=10)))
-# use caffe img_norm
-img_norm_cfg = dict(
-    mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
 train_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
@@ -20,10 +17,7 @@ train_pipeline = [
         multiscale_mode='range',
         keep_ratio=True),
     dict(type='RandomFlip', flip_ratio=0.5),
-    dict(type='Normalize', **img_norm_cfg),
-    dict(type='Pad', size_divisor=32),
-    dict(type='DefaultFormatBundle'),
-    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
+    dict(type='PackDetInputs'),
 ]
 test_pipeline = [
     dict(type='LoadImageFromFile'),
@@ -34,11 +28,11 @@ test_pipeline = [
         transforms=[
             dict(type='Resize', keep_ratio=True),
             dict(type='RandomFlip'),
-            dict(type='Normalize', **img_norm_cfg),
-            dict(type='Pad', size_divisor=32),
-            dict(type='ImageToTensor', keys=['img']),
-            dict(type='Collect', keys=['img']),
-        ])
+        ]),
+    dict(
+        type='PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor')),
 ]
 data = dict(
     train=dict(pipeline=train_pipeline),

configs/nuimages/mask-rcnn_r50_caffe_fpn_coco-3x_20e_nuim.py

@@ -8,9 +8,6 @@ model = dict(
     backbone=dict(norm_cfg=dict(requires_grad=False), style='caffe'),
     roi_head=dict(
         bbox_head=dict(num_classes=10), mask_head=dict(num_classes=10)))
-# use caffe img_norm
-img_norm_cfg = dict(
-    mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
 train_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
@@ -20,10 +17,7 @@ train_pipeline = [
         multiscale_mode='range',
         keep_ratio=True),
     dict(type='RandomFlip', flip_ratio=0.5),
-    dict(type='Normalize', **img_norm_cfg),
-    dict(type='Pad', size_divisor=32),
-    dict(type='DefaultFormatBundle'),
-    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
+    dict(type='PackDetInputs'),
 ]
 test_pipeline = [
     dict(type='LoadImageFromFile'),
@@ -34,11 +28,11 @@ test_pipeline = [
         transforms=[
             dict(type='Resize', keep_ratio=True),
             dict(type='RandomFlip'),
-            dict(type='Normalize', **img_norm_cfg),
-            dict(type='Pad', size_divisor=32),
-            dict(type='ImageToTensor', keys=['img']),
-            dict(type='Collect', keys=['img']),
-        ])
+        ]),
+    dict(
+        type='PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor')),
 ]
 data = dict(
     train=dict(pipeline=train_pipeline),

configs/nuimages/mask-rcnn_r50_fpn_coco-2x_1x_nus-2d.py

@@ -13,8 +13,6 @@ file_client_args = dict(
         './data/nuscenes/': 's3://nuscenes/nuscenes/',
         'data/nuscenes/': 's3://nuscenes/nuscenes/'
     }))
-img_norm_cfg = dict(
-    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
 
 test_pipeline = [
     dict(type='LoadImageFromFile'),
@@ -25,11 +23,11 @@ test_pipeline = [
         transforms=[
             dict(type='Resize', keep_ratio=True),
             dict(type='RandomFlip'),
-            dict(type='Normalize', **img_norm_cfg),
-            dict(type='Pad', size_divisor=32),
-            dict(type='ImageToTensor', keys=['img']),
-            dict(type='Collect', keys=['img']),
-        ])
+        ]),
+    dict(
+        type='PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor')),
 ]
 data_root = 'data/nuimages/'
 # data = dict(

configs/parta2/PartA2_hv_secfpn_8xb2-cyclic-80e_kitti-3d-car.py

@@ -124,12 +124,8 @@ test_pipeline = [
             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'])
-        ])
+        ]),
+    dict(type='Pack3DDetInputs', keys=['points'])
 ]
 
 train_dataloader = dict(

configs/pgd/pgd_r101-caffe_fpn_head-gn_16xb2-1x_nus-mono3d.py

@@ -44,8 +44,6 @@ class_names = [
     'car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle',
     'motorcycle', 'pedestrian', 'traffic_cone', 'barrier'
 ]
-img_norm_cfg = dict(
-    mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], to_rgb=False)
 train_pipeline = [
     dict(type='LoadImageFromFileMono3D'),
     dict(
@@ -58,11 +56,8 @@ train_pipeline = [
         with_bbox_depth=True),
     dict(type='Resize', img_scale=(1600, 900), keep_ratio=True),
     dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5),
-    dict(type='Normalize', **img_norm_cfg),
-    dict(type='Pad', size_divisor=32),
-    dict(type='DefaultFormatBundle3D', class_names=class_names),
     dict(
-        type='Collect3D',
+        type='Pack3DDetInputs',
         keys=[
             'img', 'gt_bboxes', 'gt_bboxes_labels', 'attr_labels',
             'gt_bboxes_3d', 'gt_labels_3d', 'centers2d', 'depths'
@@ -76,14 +71,8 @@ test_pipeline = [
         flip=False,
         transforms=[
             dict(type='RandomFlip3D'),
-            dict(type='Normalize', **img_norm_cfg),
-            dict(type='Pad', size_divisor=32),
-            dict(
-                type='DefaultFormatBundle3D',
-                class_names=class_names,
-                with_label=False),
-            dict(type='Collect3D', keys=['img']),
-        ])
+        ]),
+    dict(type='Pack3DDetInputs', keys=['img']),
 ]
 data = dict(
     samples_per_gpu=2,

tools/dataset_converters/update_infos_to_v2.py

@@ -284,8 +284,8 @@ def update_nuscenes_infos(pkl_path, out_dir):
             ori_info_dict['ego2global_translation'])
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict.get(
             'num_features', 5)
-        temp_data_info['lidar_points']['lidar_path'] = Path(ori_info_dict[
-            'lidar_path']).name
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['lidar_path']).name
         temp_data_info['lidar_points'][
             'lidar2ego'] = convert_quaternion_to_matrix(
                 ori_info_dict['lidar2ego_rotation'],
@@ -315,8 +315,8 @@ def update_nuscenes_infos(pkl_path, out_dir):
         temp_data_info['images'] = {}
         for cam in ori_info_dict['cams']:
             empty_img_info = get_empty_img_info()
-            empty_img_info['img_path'] = Path(ori_info_dict['cams'][cam][
-                'data_path']).name
+            empty_img_info['img_path'] = Path(
+                ori_info_dict['cams'][cam]['data_path']).name
             empty_img_info['cam2img'] = ori_info_dict['cams'][cam][
                 'cam_intrinsic'].tolist()
             empty_img_info['sample_data_token'] = ori_info_dict['cams'][cam][
@@ -411,15 +411,15 @@ def update_kitti_infos(pkl_path, out_dir):
         temp_data_info['images']['CAM3']['cam2img'] = ori_info_dict['calib'][
             'P3'].tolist()
 
-        temp_data_info['images']['CAM2']['img_path'] = Path(ori_info_dict['image'][
-            'image_path']).name
+        temp_data_info['images']['CAM2']['img_path'] = Path(
+            ori_info_dict['image']['image_path']).name
         h, w = ori_info_dict['image']['image_shape']
         temp_data_info['images']['CAM2']['height'] = h
         temp_data_info['images']['CAM2']['width'] = w
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = Path(ori_info_dict[
-            'point_cloud']['velodyne_path']).name
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['point_cloud']['velodyne_path']).name
 
         rect = ori_info_dict['calib']['R0_rect'].astype(np.float32)
         Trv2c = ori_info_dict['calib']['Tr_velo_to_cam'].astype(np.float32)
@@ -533,12 +533,12 @@ def update_s3dis_infos(pkl_path, out_dir):
         temp_data_info['sample_idx'] = i
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = Path(ori_info_dict[
-            'pts_path']).name
-        temp_data_info['pts_semantic_mask_path'] = Path(ori_info_dict[
-            'pts_semantic_mask_path']).name
-        temp_data_info['pts_instance_mask_path'] = Path(ori_info_dict[
-            'pts_instance_mask_path']).name
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['pts_path']).name
+        temp_data_info['pts_semantic_mask_path'] = Path(
+            ori_info_dict['pts_semantic_mask_path']).name
+        temp_data_info['pts_instance_mask_path'] = Path(
+            ori_info_dict['pts_instance_mask_path']).name
 
         # TODO support camera
         # np.linalg.inv(info['axis_align_matrix'] @ extrinsic): depth2cam
@@ -607,12 +607,12 @@ def update_scannet_infos(pkl_path, out_dir):
         temp_data_info = get_empty_standard_data_info()
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = Path(ori_info_dict[
-            'pts_path']).name
-        temp_data_info['pts_semantic_mask_path'] = Path(ori_info_dict[
-            'pts_semantic_mask_path']).name
-        temp_data_info['pts_instance_mask_path'] = Path(ori_info_dict[
-            'pts_instance_mask_path']).name
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['pts_path']).name
+        temp_data_info['pts_semantic_mask_path'] = Path(
+            ori_info_dict['pts_semantic_mask_path']).name
+        temp_data_info['pts_instance_mask_path'] = Path(
+            ori_info_dict['pts_instance_mask_path']).name
 
         # TODO support camera
         # np.linalg.inv(info['axis_align_matrix'] @ extrinsic): depth2cam
@@ -679,8 +679,8 @@ def update_sunrgbd_infos(pkl_path, out_dir):
         temp_data_info = get_empty_standard_data_info()
         temp_data_info['lidar_points']['num_pts_feats'] = ori_info_dict[
             'point_cloud']['num_features']
-        temp_data_info['lidar_points']['lidar_path'] = Path(ori_info_dict[
-            'pts_path']).name
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['pts_path']).name
         calib = ori_info_dict['calib']
         rt_mat = calib['Rt']
         # follow Coord3DMode.convert_point
@@ -688,8 +688,8 @@ def update_sunrgbd_infos(pkl_path, out_dir):
                            ]) @ rt_mat.transpose(1, 0)
         depth2img = calib['K'] @ rt_mat
         temp_data_info['images']['CAM0']['depth2img'] = depth2img.tolist()
-        temp_data_info['images']['CAM0']['img_path'] = Path(ori_info_dict['image'][
-            'image_path']).name
+        temp_data_info['images']['CAM0']['img_path'] = Path(
+            ori_info_dict['image']['image_path']).name
         h, w = ori_info_dict['image']['image_shape']
         temp_data_info['images']['CAM0']['height'] = h
         temp_data_info['images']['CAM0']['width'] = w
@@ -761,8 +761,8 @@ def update_lyft_infos(pkl_path, out_dir):
         temp_data_info['ego2global'] = convert_quaternion_to_matrix(
             ori_info_dict['ego2global_rotation'],
             ori_info_dict['ego2global_translation'])
-        temp_data_info['lidar_points']['lidar_path'] = Path(ori_info_dict[
-            'lidar_path']).name
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['lidar_path']).name
         temp_data_info['lidar_points'][
             'lidar2ego'] = convert_quaternion_to_matrix(
                 ori_info_dict['lidar2ego_rotation'],
@@ -793,8 +793,8 @@ def update_lyft_infos(pkl_path, out_dir):
         temp_data_info['images'] = {}
         for cam in ori_info_dict['cams']:
             empty_img_info = get_empty_img_info()
-            empty_img_info['img_path'] = Path(ori_info_dict['cams'][cam][
-                'data_path']).name
+            empty_img_info['img_path'] = Path(
+                ori_info_dict['cams'][cam]['data_path']).name
             empty_img_info['cam2img'] = ori_info_dict['cams'][cam][
                 'cam_intrinsic'].tolist()
             empty_img_info['sample_data_token'] = ori_info_dict['cams'][cam][
@@ -913,8 +913,8 @@ def update_waymo_infos(pkl_path, out_dir):
             'point_cloud']['num_features']
         temp_data_info['lidar_points']['timestamp'] = ori_info_dict[
             'timestamp']
-        temp_data_info['lidar_points']['lidar_path'] = Path(ori_info_dict[
-            'point_cloud']['velodyne_path']).name
+        temp_data_info['lidar_points']['lidar_path'] = Path(
+            ori_info_dict['point_cloud']['velodyne_path']).name
 
         # TODO discuss the usage of Tr_velo_to_cam in lidar
         Trv2c = ori_info_dict['calib']['Tr_velo_to_cam'].astype(np.float32)
@@ -934,8 +934,8 @@ def update_waymo_infos(pkl_path, out_dir):
             lidar_sweep = get_single_lidar_sweep()
             lidar_sweep['ego2global'] = ori_sweep['pose']
             lidar_sweep['timestamp'] = ori_sweep['timestamp']
-            lidar_sweep['lidar_points']['lidar_path'] = Path(ori_sweep[
-                'velodyne_path']).name
+            lidar_sweep['lidar_points']['lidar_path'] = Path(
+                ori_sweep['velodyne_path']).name
             # image sweeps
             image_sweep = get_single_image_sweep(camera_types)
             image_sweep['ego2global'] = ori_sweep['pose']