[Feature] File I/O migration and reconstruction (#2319)

* refactor leio

* update mmengine and mmcv version

* update

* update docs

* update version

mmdet3d/evaluation/functional/waymo_utils/prediction_to_waymo.py

@@ -41,8 +41,8 @@ class Prediction2Waymo(object):
             validation and 2 for testing.
         classes (dict): A list of class name.
         workers (str): Number of parallel processes. Defaults to 2.
-        file_client_args (str): File client for reading gt in waymo format.
-            Defaults to ``dict(backend='disk')``.
+        backend_args (dict, optional): Arguments to instantiate the
+            corresponding backend. Defaults to None.
         from_kitti_format (bool, optional): Whether the reuslts are kitti
             format. Defaults to False.
         idx2metainfo (Optional[dict], optional): The mapping from sample_idx to
@@ -58,7 +58,7 @@ class Prediction2Waymo(object):
                  prefix: str,
                  classes: dict,
                  workers: int = 2,
-                 file_client_args: dict = dict(backend='disk'),
+                 backend_args: Optional[dict] = None,
                  from_kitti_format: bool = False,
                  idx2metainfo: Optional[dict] = None):
 
@@ -69,7 +69,7 @@ class Prediction2Waymo(object):
         self.prefix = prefix
         self.classes = classes
         self.workers = int(workers)
-        self.file_client_args = file_client_args
+        self.backend_args = backend_args
         self.from_kitti_format = from_kitti_format
         if idx2metainfo is not None:
             self.idx2metainfo = idx2metainfo
@@ -114,12 +114,12 @@ class Prediction2Waymo(object):
 
     def get_file_names(self):
         """Get file names of waymo raw data."""
-        if 'path_mapping' in self.file_client_args:
-            for path in self.file_client_args['path_mapping'].keys():
+        if 'path_mapping' in self.backend_args:
+            for path in self.backend_args['path_mapping'].keys():
                 if path in self.waymo_tfrecords_dir:
                     self.waymo_tfrecords_dir = \
                         self.waymo_tfrecords_dir.replace(
-                            path, self.file_client_args['path_mapping'][path])
+                            path, self.backend_args['path_mapping'][path])
             from petrel_client.client import Client
             client = Client()
             contents = client.list(self.waymo_tfrecords_dir)

mmdet3d/evaluation/metrics/kitti_metric.py

@@ -47,24 +47,21 @@ class KittiMetric(BaseMetric):
         collect_device (str): Device name used for collecting results
             from different ranks during distributed training. Must be 'cpu' or
             'gpu'. Defaults to 'cpu'.
-        file_client_args (dict): Arguments to instantiate a FileClient.
-            See :class:`mmengine.fileio.FileClient` for details.
-            Defaults to dict(backend='disk').
+        backend_args (dict, optional): Arguments to instantiate the
+            corresponding backend. Defaults to None.
     """
 
-    def __init__(
-        self,
-        ann_file: str,
-        metric: Union[str, List[str]] = 'bbox',
-        pcd_limit_range: List[float] = [0, -40, -3, 70.4, 40, 0.0],
-        prefix: Optional[str] = None,
-        pklfile_prefix: Optional[str] = None,
-        default_cam_key: str = 'CAM2',
-        format_only: bool = False,
-        submission_prefix: Optional[str] = None,
-        collect_device: str = 'cpu',
-        file_client_args: dict = dict(backend='disk')
-    ) -> None:
+    def __init__(self,
+                 ann_file: str,
+                 metric: Union[str, List[str]] = 'bbox',
+                 pcd_limit_range: List[float] = [0, -40, -3, 70.4, 40, 0.0],
+                 prefix: Optional[str] = None,
+                 pklfile_prefix: Optional[str] = None,
+                 default_cam_key: str = 'CAM2',
+                 format_only: bool = False,
+                 submission_prefix: Optional[str] = None,
+                 collect_device: str = 'cpu',
+                 backend_args: Optional[dict] = None) -> None:
         self.default_prefix = 'Kitti metric'
         super(KittiMetric, self).__init__(
             collect_device=collect_device, prefix=prefix)
@@ -80,7 +77,7 @@ class KittiMetric(BaseMetric):
 
         self.submission_prefix = submission_prefix
         self.default_cam_key = default_cam_key
-        self.file_client_args = file_client_args
+        self.backend_args = backend_args
 
         allowed_metrics = ['bbox', 'img_bbox', 'mAP', 'LET_mAP']
         self.metrics = metric if isinstance(metric, list) else [metric]
@@ -188,7 +185,7 @@ class KittiMetric(BaseMetric):
         self.classes = self.dataset_meta['classes']
 
         # load annotations
-        pkl_infos = load(self.ann_file, file_client_args=self.file_client_args)
+        pkl_infos = load(self.ann_file, backend_args=self.backend_args)
         self.data_infos = self.convert_annos_to_kitti_annos(pkl_infos)
         result_dict, tmp_dir = self.format_results(
             results,

mmdet3d/evaluation/metrics/lyft_metric.py

@@ -46,21 +46,19 @@ class LyftMetric(BaseMetric):
             'gpu'. Defaults to 'cpu'.
     """
 
-    def __init__(
-        self,
-        data_root: str,
-        ann_file: str,
-        metric: Union[str, List[str]] = 'bbox',
-        modality=dict(
-            use_camera=False,
-            use_lidar=True,
-        ),
-        prefix: Optional[str] = None,
-        jsonfile_prefix: str = None,
-        csv_savepath: str = None,
-        collect_device: str = 'cpu',
-        file_client_args: dict = dict(backend='disk')
-    ) -> None:
+    def __init__(self,
+                 data_root: str,
+                 ann_file: str,
+                 metric: Union[str, List[str]] = 'bbox',
+                 modality=dict(
+                     use_camera=False,
+                     use_lidar=True,
+                 ),
+                 prefix: Optional[str] = None,
+                 jsonfile_prefix: str = None,
+                 csv_savepath: str = None,
+                 collect_device: str = 'cpu',
+                 backend_args: Optional[dict] = None) -> None:
         self.default_prefix = 'Lyft metric'
         super(LyftMetric, self).__init__(
             collect_device=collect_device, prefix=prefix)
@@ -68,7 +66,7 @@ class LyftMetric(BaseMetric):
         self.data_root = data_root
         self.modality = modality
         self.jsonfile_prefix = jsonfile_prefix
-        self.file_client_args = file_client_args
+        self.backend_args = backend_args
         self.csv_savepath = csv_savepath
         self.metrics = metric if isinstance(metric, list) else [metric]
 
@@ -115,7 +113,7 @@ class LyftMetric(BaseMetric):
         # load annotations
 
         self.data_infos = load(
-            self.ann_file, file_client_args=self.file_client_args)['data_list']
+            self.ann_file, backend_args=self.backend_args)['data_list']
         result_dict, tmp_dir = self.format_results(results, classes,
                                                    self.jsonfile_prefix)
 

mmdet3d/evaluation/metrics/nuscenes_metric.py

@@ -47,9 +47,8 @@ class NuScenesMetric(BaseMetric):
         collect_device (str): Device name used for collecting results
             from different ranks during distributed training. Must be 'cpu' or
             'gpu'. Defaults to 'cpu'.
-        file_client_args (dict): Arguments to instantiate a FileClient.
-            See :class:`mmengine.fileio.FileClient` for details.
-            Defaults to dict(backend='disk').
+        backend_args (dict, optional): Arguments to instantiate the
+            corresponding backend. Defaults to None.
     """
     NameMapping = {
         'movable_object.barrier': 'barrier',
@@ -88,19 +87,17 @@ class NuScenesMetric(BaseMetric):
         'attr_err': 'mAAE'
     }
 
-    def __init__(
-        self,
-        data_root: str,
-        ann_file: str,
-        metric: Union[str, List[str]] = 'bbox',
-        modality: dict = dict(use_camera=False, use_lidar=True),
-        prefix: Optional[str] = None,
-        format_only: bool = False,
-        jsonfile_prefix: Optional[str] = None,
-        eval_version: str = 'detection_cvpr_2019',
-        collect_device: str = 'cpu',
-        file_client_args: dict = dict(backend='disk')
-    ) -> None:
+    def __init__(self,
+                 data_root: str,
+                 ann_file: str,
+                 metric: Union[str, List[str]] = 'bbox',
+                 modality: dict = dict(use_camera=False, use_lidar=True),
+                 prefix: Optional[str] = None,
+                 format_only: bool = False,
+                 jsonfile_prefix: Optional[str] = None,
+                 eval_version: str = 'detection_cvpr_2019',
+                 collect_device: str = 'cpu',
+                 backend_args: Optional[dict] = None) -> None:
         self.default_prefix = 'NuScenes metric'
         super(NuScenesMetric, self).__init__(
             collect_device=collect_device, prefix=prefix)
@@ -120,7 +117,7 @@ class NuScenesMetric(BaseMetric):
             'the end.'
 
         self.jsonfile_prefix = jsonfile_prefix
-        self.file_client_args = file_client_args
+        self.backend_args = backend_args
 
         self.metrics = metric if isinstance(metric, list) else [metric]
 
@@ -169,7 +166,7 @@ class NuScenesMetric(BaseMetric):
         self.version = self.dataset_meta['version']
         # load annotations
         self.data_infos = load(
-            self.ann_file, file_client_args=self.file_client_args)['data_list']
+            self.ann_file, backend_args=self.backend_args)['data_list']
         result_dict, tmp_dir = self.format_results(results, classes,
                                                    self.jsonfile_prefix)
 

mmdet3d/evaluation/metrics/waymo_metric.py

@@ -69,8 +69,8 @@ class WaymoMetric(KittiMetric):
         collect_device (str): Device name used for collecting results
             from different ranks during distributed training. Must be 'cpu' or
             'gpu'. Defaults to 'cpu'.
-        file_client_args (dict): File client for reading gt in waymo format.
-            Defaults to ``dict(backend='disk')``.
+        backend_args (dict, optional): Arguments to instantiate the
+            corresponding backend. Defaults to None.
         idx2metainfo (str, optional): The file path of the metainfo in waymo.
             It stores the mapping from sample_idx to metainfo. The metainfo
             must contain the keys: 'idx2contextname' and 'idx2timestamp'.
@@ -94,7 +94,7 @@ class WaymoMetric(KittiMetric):
                  default_cam_key: str = 'CAM_FRONT',
                  use_pred_sample_idx: bool = False,
                  collect_device: str = 'cpu',
-                 file_client_args: dict = dict(backend='disk'),
+                 backend_args: Optional[dict] = None,
                  idx2metainfo: Optional[str] = None) -> None:
         self.waymo_bin_file = waymo_bin_file
         self.data_root = data_root
@@ -117,7 +117,7 @@ class WaymoMetric(KittiMetric):
             submission_prefix=submission_prefix,
             default_cam_key=default_cam_key,
             collect_device=collect_device,
-            file_client_args=file_client_args)
+            backend_args=backend_args)
         self.format_only = format_only
         if self.format_only:
             assert pklfile_prefix is not None, 'pklfile_prefix must be '
@@ -388,7 +388,7 @@ class WaymoMetric(KittiMetric):
             waymo_results_final_path,
             prefix,
             classes,
-            file_client_args=self.file_client_args,
+            backend_args=self.backend_args,
             from_kitti_format=self.convert_kitti_format,
             idx2metainfo=self.idx2metainfo)
         converter.convert()

mmdet3d/utils/misc.py

@@ -5,7 +5,7 @@ def replace_ceph_backend(cfg):
     cfg_pretty_text = cfg.pretty_text
 
     replace_strs = \
-        r'''file_client_args = dict(
+        r'''backend_args = dict(
             backend='petrel',
             path_mapping=dict({
                 './data/DATA/': 's3://openmmlab/datasets/detection3d/CEPH/',
@@ -49,6 +49,8 @@ def replace_ceph_backend(cfg):
     # cfg_pretty_text = cfg_pretty_text.replace(
     #   'ann_file', replace_strs + ', ann_file')
 
+    cfg_pretty_text = cfg_pretty_text.replace('backend_args=None', '')
+
     # replace LoadImageFromFile
     cfg_pretty_text = cfg_pretty_text.replace(
         'LoadImageFromFile\'', 'LoadImageFromFile\',' + replace_strs)
@@ -80,6 +82,18 @@ def replace_ceph_backend(cfg):
     cfg_pretty_text = cfg_pretty_text.replace(
         'LoadAnnotations3D\'', 'LoadAnnotations3D\',' + replace_strs)
 
+    # replace KittiMetric
+    cfg_pretty_text = cfg_pretty_text.replace('KittiMetric\'',
+                                              'KittiMetric\',' + replace_strs)
+
+    # replace LyftMetric
+    cfg_pretty_text = cfg_pretty_text.replace('LyftMetric\'',
+                                              'LyftMetric\',' + replace_strs)
+
+    # replace NuScenesMetric
+    cfg_pretty_text = cfg_pretty_text.replace(
+        'NuScenesMetric\'', 'NuScenesMetric\',' + replace_strs)
+
     # replace WaymoMetric
     cfg_pretty_text = cfg_pretty_text.replace('WaymoMetric\'',
                                               'WaymoMetric\',' + replace_strs)

projects/BEVFusion/bevfusion/loading.py

@@ -3,8 +3,8 @@ import copy
 from typing import Optional
 
 import mmcv
-import mmengine
 import numpy as np
+from mmengine.fileio import get
 
 from mmdet3d.datasets.transforms import LoadMultiViewImageFromFiles
 from mmdet3d.registry import TRANSFORMS
@@ -23,9 +23,8 @@ class BEVLoadMultiViewImageFromFiles(LoadMultiViewImageFromFiles):
         to_float32 (bool): Whether to convert the img to float32.
             Defaults to False.
         color_type (str): Color type of the file. Defaults to 'unchanged'.
-        file_client_args (dict): Arguments to instantiate a FileClient.
-            See :class:`mmengine.fileio.FileClient` for details.
-            Defaults to dict(backend='disk').
+        backend_args (dict, optional): Arguments to instantiate the
+            corresponding backend. Defaults to None.
         num_views (int): Number of view in a frame. Defaults to 5.
         num_ref_frames (int): Number of frame in loading. Defaults to -1.
         test_mode (bool): Whether is test mode in loading. Defaults to False.
@@ -160,12 +159,11 @@ class BEVLoadMultiViewImageFromFiles(LoadMultiViewImageFromFiles):
 
         results['ori_cam2img'] = copy.deepcopy(results['cam2img'])
 
-        if self.file_client is None:
-            self.file_client = mmengine.FileClient(**self.file_client_args)
-
         # img is of shape (h, w, c, num_views)
         # h and w can be different for different views
-        img_bytes = [self.file_client.get(name) for name in filename]
+        img_bytes = [
+            get(name, backend_args=self.backend_args) for name in filename
+        ]
         imgs = [
             mmcv.imfrombytes(
                 img_byte,

projects/BEVFusion/configs/bevfusion_voxel0075_second_secfpn_8xb4-cyclic-20e_nus-3d.py

@@ -27,7 +27,7 @@ data_prefix = dict(
     CAM_BACK_LEFT='samples/CAM_BACK_LEFT',
     sweeps='sweeps/LIDAR_TOP')
 input_modality = dict(use_lidar=True, use_camera=True)
-file_client_args = dict(backend='disk')
+backend_args = None
 
 model = dict(
     type='BEVFusion',
@@ -209,20 +209,24 @@ db_sampler = dict(
         coord_type='LIDAR',
         load_dim=5,
         use_dim=[0, 1, 2, 3, 4],
-        reduce_beams=32))
+        reduce_beams=32,
+        backend_args=backend_args),
+    backend_args=backend_args)
 
 train_pipeline = [
     dict(
         type='BEVLoadMultiViewImageFromFiles',
         to_float32=True,
-        color_type='color'),
+        color_type='color',
+        backend_args=backend_args),
     dict(
         type='LoadPointsFromFile',
         coord_type='LIDAR',
         load_dim=5,
         use_dim=5,
         reduce_beams=32,
-        load_augmented=None),
+        load_augmented=None,
+        backend_args=backend_args),
     dict(
         type='LoadPointsFromMultiSweeps',
         sweeps_num=9,
@@ -231,7 +235,8 @@ train_pipeline = [
         reduce_beams=32,
         pad_empty_sweeps=True,
         remove_close=True,
-        load_augmented=None),
+        load_augmented=None,
+        backend_args=backend_args),
     dict(
         type='LoadAnnotations3D',
         with_bbox_3d=True,
@@ -285,15 +290,22 @@ test_pipeline = [
     dict(
         type='BEVLoadMultiViewImageFromFiles',
         to_float32=True,
-        color_type='color'),
-    dict(type='LoadPointsFromFile', coord_type='LIDAR', load_dim=5, use_dim=5),
+        color_type='color',
+        backend_args=backend_args),
+    dict(
+        type='LoadPointsFromFile',
+        coord_type='LIDAR',
+        load_dim=5,
+        use_dim=5,
+        backend_args=backend_args),
     dict(
         type='LoadPointsFromMultiSweeps',
         sweeps_num=9,
         load_dim=5,
         use_dim=5,
         pad_empty_sweeps=True,
-        remove_close=True),
+        remove_close=True,
+        backend_args=backend_args),
     dict(
         type='ImageAug3D',
         final_dim=[256, 704],
@@ -331,7 +343,8 @@ train_dataloader = dict(
         data_prefix=data_prefix,
         # 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'))
+        box_type_3d='LiDAR',
+        backend_args=backend_args))
 val_dataloader = dict(
     batch_size=1,
     num_workers=0,
@@ -347,14 +360,16 @@ val_dataloader = dict(
         modality=input_modality,
         data_prefix=data_prefix,
         test_mode=True,
-        box_type_3d='LiDAR'))
+        box_type_3d='LiDAR',
+        backend_args=backend_args))
 test_dataloader = val_dataloader
 
 val_evaluator = dict(
     type='NuScenesMetric',
     data_root=data_root,
     ann_file=data_root + 'nuscenes_infos_val.pkl',
-    metric='bbox')
+    metric='bbox',
+    backend_args=backend_args)
 test_evaluator = val_evaluator
 
 vis_backends = [dict(type='LocalVisBackend')]

projects/CenterFormer/configs/centerformer_voxel01_second-attn_secfpn-attn_4xb4-cyclic-20e_waymoD5-3d-3class.py

@@ -13,7 +13,7 @@ class_names = ['Car', 'Pedestrian', 'Cyclist']
 tasks = [dict(num_class=3, class_names=['car', 'pedestrian', 'cyclist'])]
 metainfo = dict(classes=class_names)
 input_modality = dict(use_lidar=True, use_camera=False)
-file_client_args = dict(backend='disk')
+backend_args = None
 
 model = dict(
     type='CenterFormer',
@@ -120,7 +120,9 @@ db_sampler = dict(
         type='LoadPointsFromFile',
         coord_type='LIDAR',
         load_dim=6,
-        use_dim=[0, 1, 2, 3, 4]))
+        use_dim=[0, 1, 2, 3, 4],
+        backend_args=backend_args),
+    backend_args=backend_args)
 
 train_pipeline = [
     dict(
@@ -128,7 +130,8 @@ train_pipeline = [
         coord_type='LIDAR',
         load_dim=6,
         use_dim=5,
-        norm_intensity=True),
+        norm_intensity=True,
+        backend_args=backend_args),
     # Add this if using `MultiFrameDeformableDecoderRPN`
     # dict(
     #     type='LoadPointsFromMultiSweeps',
@@ -160,7 +163,7 @@ test_pipeline = [
         load_dim=6,
         use_dim=5,
         norm_intensity=True,
-        file_client_args=file_client_args),
+        backend_args=backend_args),
     dict(
         type='MultiScaleFlipAug3D',
         img_scale=(1333, 800),
@@ -199,7 +202,7 @@ train_dataloader = dict(
         box_type_3d='LiDAR',
         # load one frame every five frames
         load_interval=5,
-        file_client_args=file_client_args))
+        backend_args=backend_args))
 val_dataloader = dict(
     batch_size=1,
     num_workers=1,
@@ -216,7 +219,7 @@ val_dataloader = dict(
         test_mode=True,
         metainfo=metainfo,
         box_type_3d='LiDAR',
-        file_client_args=file_client_args))
+        backend_args=backend_args))
 test_dataloader = val_dataloader
 
 val_evaluator = dict(
@@ -224,7 +227,7 @@ val_evaluator = dict(
     ann_file='./data/waymo/kitti_format/waymo_infos_val.pkl',
     waymo_bin_file='./data/waymo/waymo_format/gt.bin',
     data_root='./data/waymo/waymo_format',
-    file_client_args=file_client_args,
+    backend_args=backend_args,
     convert_kitti_format=False,
     idx2metainfo='./data/waymo/waymo_format/idx2metainfo.pkl')
 test_evaluator = val_evaluator

projects/DETR3D/configs/detr3d_r101_gridmask.py

@@ -128,9 +128,13 @@ test_transforms = [
 ]
 train_transforms = [dict(type='PhotoMetricDistortion3D')] + test_transforms
 
-file_client_args = dict(backend='disk')
+backend_args = None
 train_pipeline = [
-    dict(type='LoadMultiViewImageFromFiles', to_float32=True, num_views=6),
+    dict(
+        type='LoadMultiViewImageFromFiles',
+        to_float32=True,
+        num_views=6,
+        backend_args=backend_args),
     dict(
         type='LoadAnnotations3D',
         with_bbox_3d=True,
@@ -143,7 +147,11 @@ train_pipeline = [
 ]
 
 test_pipeline = [
-    dict(type='LoadMultiViewImageFromFiles', to_float32=True, num_views=6),
+    dict(
+        type='LoadMultiViewImageFromFiles',
+        to_float32=True,
+        num_views=6,
+        backend_args=backend_args),
     dict(type='MultiViewWrapper', transforms=test_transforms),
     dict(type='Pack3DDetInputs', keys=['img'])
 ]
@@ -176,7 +184,8 @@ train_dataloader = dict(
         data_prefix=data_prefix,
         # 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'))
+        box_type_3d='LiDAR',
+        backend_args=backend_args))
 
 val_dataloader = dict(
     batch_size=1,
@@ -194,7 +203,8 @@ val_dataloader = dict(
         modality=input_modality,
         test_mode=True,
         data_prefix=data_prefix,
-        box_type_3d='LiDAR'))
+        box_type_3d='LiDAR',
+        backend_args=backend_args))
 
 test_dataloader = val_dataloader
 
@@ -202,7 +212,8 @@ val_evaluator = dict(
     type='NuScenesMetric',
     data_root=data_root,
     ann_file=data_root + 'nuscenes_infos_val.pkl',
-    metric='bbox')
+    metric='bbox',
+    backend_args=backend_args)
 test_evaluator = val_evaluator
 
 optim_wrapper = dict(

projects/PETR/config/petr/petr_vovnet_gridmask_p4_800x320.py

@@ -115,7 +115,7 @@ model = dict(
 
 dataset_type = 'NuScenesDataset'
 data_root = 'data/nuscenes/'
-file_client_args = dict(backend='disk')
+backend_args = None
 
 db_sampler = dict(
     data_root=data_root,
@@ -150,7 +150,9 @@ db_sampler = dict(
         type='LoadPointsFromFile',
         coord_type='LIDAR',
         load_dim=5,
-        use_dim=[0, 1, 2, 3, 4]))
+        use_dim=[0, 1, 2, 3, 4],
+        backend_args=backend_args),
+    backend_args=backend_args)
 ida_aug_conf = {
     'resize_lim': (0.47, 0.625),
     'final_dim': (320, 800),
@@ -162,7 +164,10 @@ ida_aug_conf = {
 }
 
 train_pipeline = [
-    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(
+        type='LoadMultiViewImageFromFiles',
+        to_float32=True,
+        backend_args=backend_args),
     dict(
         type='LoadAnnotations3D',
         with_bbox_3d=True,
@@ -187,7 +192,10 @@ train_pipeline = [
         ])
 ]
 test_pipeline = [
-    dict(type='LoadMultiViewImageFromFiles', to_float32=True),
+    dict(
+        type='LoadMultiViewImageFromFiles',
+        to_float32=True,
+        backend_args=backend_args),
     dict(
         type='ResizeCropFlipImage', data_aug_conf=ida_aug_conf,
         training=False),
@@ -212,7 +220,8 @@ train_dataloader = dict(
         metainfo=metainfo,
         test_mode=False,
         modality=input_modality,
-        use_valid_flag=True))
+        use_valid_flag=True,
+        backend_args=backend_args))
 test_dataloader = dict(
     dataset=dict(
         type=dataset_type,
@@ -229,7 +238,8 @@ test_dataloader = dict(
         metainfo=metainfo,
         test_mode=True,
         modality=input_modality,
-        use_valid_flag=True))
+        use_valid_flag=True,
+        backend_args=backend_args))
 val_dataloader = dict(
     dataset=dict(
         type=dataset_type,
@@ -246,7 +256,8 @@ val_dataloader = dict(
         metainfo=metainfo,
         test_mode=True,
         modality=input_modality,
-        use_valid_flag=True))
+        use_valid_flag=True,
+        backend_args=backend_args))
 
 # Different from original PETR:
 # We don't use special lr for image_backbone

requirements/mminstall.txt

 mmcv>=2.0.0rc4,<2.1.0
 mmdet>=3.0.0rc0,<3.1.0
-mmengine>=0.4.0,<1.0.0
+mmengine>=0.6.0,<1.0.0

requirements/readthedocs.txt

-mmcv>=2.0.0rc0
+mmcv>=2.0.0rc4
 mmdet>=3.0.0rc0
-mmengine>=0.1.0
+mmengine>=0.6.0
 torch
 torchvision

tests/test_datasets/test_transforms/test_loading.py

@@ -16,12 +16,12 @@ class TestLoadPointsFromFile(unittest.TestCase):
 
     def test_load_points_from_file(self):
         use_dim = 3
-        file_client_args = dict(backend='disk')
+        backend_args = None
         load_points_transform = LoadPointsFromFile(
             coord_type='LIDAR',
             load_dim=4,
             use_dim=use_dim,
-            file_client_args=file_client_args)
+            backend_args=backend_args)
         data_info = create_dummy_data_info()
         info = load_points_transform(data_info)
         self.assertIn('points', info)
@@ -30,7 +30,7 @@ class TestLoadPointsFromFile(unittest.TestCase):
             coord_type='DEPTH',
             load_dim=4,
             use_dim=use_dim,
-            file_client_args=file_client_args)
+            backend_args=backend_args)
         info = load_points_transform(data_info)
         self.assertIsInstance(info['points'], DepthPoints)
         self.assertEqual(info['points'].shape[-1], use_dim)
@@ -39,7 +39,7 @@ class TestLoadPointsFromFile(unittest.TestCase):
             load_dim=4,
             use_dim=use_dim,
             shift_height=True,
-            file_client_args=file_client_args)
+            backend_args=backend_args)
         info = load_points_transform(data_info)
         # extra height dim
         self.assertEqual(info['points'].shape[-1], use_dim + 1)
@@ -53,7 +53,7 @@ class TestLoadPointsFromFile(unittest.TestCase):
 class TestLoadAnnotations3D(unittest.TestCase):
 
     def test_load_points_from_file(self):
-        file_client_args = dict(backend='disk')
+        backend_args = None
 
         load_anns_transform = LoadAnnotations3D(
             with_bbox_3d=True,
@@ -62,7 +62,7 @@ class TestLoadAnnotations3D(unittest.TestCase):
             seg_offset=2**16,
             dataset_type='semantickitti',
             seg_3d_dtype='np.uint32',
-            file_client_args=file_client_args)
+            backend_args=backend_args)
         self.assertIs(load_anns_transform.with_seg, False)
         self.assertIs(load_anns_transform.with_bbox_3d, True)
         self.assertIs(load_anns_transform.with_label_3d, True)

tools/dataset_converters/create_gt_database.py

@@ -147,7 +147,7 @@ def create_groundtruth_database(dataset_class_name,
     dataset_cfg = dict(
         type=dataset_class_name, data_root=data_path, ann_file=info_path)
     if dataset_class_name == 'KittiDataset':
-        file_client_args = dict(backend='disk')
+        backend_args = None
         dataset_cfg.update(
             modality=dict(
                 use_lidar=True,
@@ -161,12 +161,12 @@ def create_groundtruth_database(dataset_class_name,
                     coord_type='LIDAR',
                     load_dim=4,
                     use_dim=4,
-                    file_client_args=file_client_args),
+                    backend_args=backend_args),
                 dict(
                     type='LoadAnnotations3D',
                     with_bbox_3d=True,
                     with_label_3d=True,
-                    file_client_args=file_client_args)
+                    backend_args=backend_args)
             ])
 
     elif dataset_class_name == 'NuScenesDataset':
@@ -193,7 +193,7 @@ def create_groundtruth_database(dataset_class_name,
             ])
 
     elif dataset_class_name == 'WaymoDataset':
-        file_client_args = dict(backend='disk')
+        backend_args = None
         dataset_cfg.update(
             test_mode=False,
             data_prefix=dict(
@@ -210,12 +210,12 @@ def create_groundtruth_database(dataset_class_name,
                     coord_type='LIDAR',
                     load_dim=6,
                     use_dim=6,
-                    file_client_args=file_client_args),
+                    backend_args=backend_args),
                 dict(
                     type='LoadAnnotations3D',
                     with_bbox_3d=True,
                     with_label_3d=True,
-                    file_client_args=file_client_args)
+                    backend_args=backend_args)
             ])
 
     dataset = DATASETS.build(dataset_cfg)
@@ -510,7 +510,7 @@ class GTDatabaseCreater:
             data_root=self.data_path,
             ann_file=self.info_path)
         if self.dataset_class_name == 'KittiDataset':
-            file_client_args = dict(backend='disk')
+            backend_args = None
             dataset_cfg.update(
                 test_mode=False,
                 data_prefix=dict(
@@ -527,12 +527,12 @@ class GTDatabaseCreater:
                         coord_type='LIDAR',
                         load_dim=4,
                         use_dim=4,
-                        file_client_args=file_client_args),
+                        backend_args=backend_args),
                     dict(
                         type='LoadAnnotations3D',
                         with_bbox_3d=True,
                         with_label_3d=True,
-                        file_client_args=file_client_args)
+                        backend_args=backend_args)
                 ])
 
         elif self.dataset_class_name == 'NuScenesDataset':
@@ -560,7 +560,7 @@ class GTDatabaseCreater:
                 ])
 
         elif self.dataset_class_name == 'WaymoDataset':
-            file_client_args = dict(backend='disk')
+            backend_args = None
             dataset_cfg.update(
                 test_mode=False,
                 data_prefix=dict(
@@ -579,12 +579,12 @@ class GTDatabaseCreater:
                         coord_type='LIDAR',
                         load_dim=6,
                         use_dim=6,
-                        file_client_args=file_client_args),
+                        backend_args=backend_args),
                     dict(
                         type='LoadAnnotations3D',
                         with_bbox_3d=True,
                         with_label_3d=True,
-                        file_client_args=file_client_args)
+                        backend_args=backend_args)
                 ])
 
         self.dataset = DATASETS.build(dataset_cfg)