Tutorial 3: Customize Data Pipelines

Design of Data pipelines

Following typical conventions, we use Dataset and DataLoader for data loading with multiple workers. Dataset returns a dict of data items corresponding the arguments of models’ forward method. Since the data in object detection may not be the same size (image size, gt bbox size, etc.), we introduce a new DataContainer type in MMCV to help collect and distribute data of different size. See here for more details.

The data preparation pipeline and the dataset is decomposed. Usually a dataset defines how to process the annotations and a data pipeline defines all the steps to prepare a data dict. A pipeline consists of a sequence of operations. Each operation takes a dict as input and also output a dict for the next transform.

We present a classical pipeline in the following figure. The blue blocks are pipeline operations. With the pipeline going on, each operator can add new keys (marked as green) to the result dict or update the existing keys (marked as orange). ../_images/data_pipeline.pngpipeline figure

The operations are categorized into data loading, pre-processing, formatting and test-time augmentation.

Here is an pipeline example for PointPillars.

train_pipeline = [
    dict(
        type='LoadPointsFromFile',
        load_dim=5,
        use_dim=5,
        file_client_args=file_client_args),
    dict(
        type='LoadPointsFromMultiSweeps',
        sweeps_num=10,
        file_client_args=file_client_args),
    dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
    dict(
        type='GlobalRotScaleTrans',
        rot_range=[-0.3925, 0.3925],
        scale_ratio_range=[0.95, 1.05],
        translation_std=[0, 0, 0]),
    dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5),
    dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
    dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
    dict(type='ObjectNameFilter', classes=class_names),
    dict(type='PointShuffle'),
    dict(type='DefaultFormatBundle3D', class_names=class_names),
    dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
]
test_pipeline = [
    dict(
        type='LoadPointsFromFile',
        load_dim=5,
        use_dim=5,
        file_client_args=file_client_args),
    dict(
        type='LoadPointsFromMultiSweeps',
        sweeps_num=10,
        file_client_args=file_client_args),
    dict(
        type='MultiScaleFlipAug',
        img_scale=(1333, 800),
        pts_scale_ratio=1.0,
        flip=False,
        pcd_horizontal_flip=False,
        pcd_vertical_flip=False,
        transforms=[
            dict(
                type='GlobalRotScaleTrans',
                rot_range=[0, 0],
                scale_ratio_range=[1., 1.],
                translation_std=[0, 0, 0]),
            dict(type='RandomFlip3D'),
            dict(
                type='PointsRangeFilter', point_cloud_range=point_cloud_range),
            dict(
                type='DefaultFormatBundle3D',
                class_names=class_names,
                with_label=False),
            dict(type='Collect3D', keys=['points'])
        ])
]

For each operation, we list the related dict fields that are added/updated/removed.

Data loading

LoadPointsFromFile

  • add: points

LoadPointsFromMultiSweeps

  • update: points

LoadAnnotations3D

  • add: gt_bboxes_3d, gt_labels_3d, gt_bboxes, gt_labels, pts_instance_mask, pts_semantic_mask, bbox3d_fields, pts_mask_fields, pts_seg_fields

Pre-processing

GlobalRotScaleTrans

  • add: pcd_trans, pcd_rotation, pcd_scale_factor

  • update: points, *bbox3d_fields

RandomFlip3D

  • add: flip, pcd_horizontal_flip, pcd_vertical_flip

  • update: points, *bbox3d_fields

PointsRangeFilter

  • update: points

ObjectRangeFilter

  • update: gt_bboxes_3d, gt_labels_3d

ObjectNameFilter

  • update: gt_bboxes_3d, gt_labels_3d

PointShuffle

  • update: points

PointsRangeFilter

  • update: points

Formatting

DefaultFormatBundle3D

  • update: points, gt_bboxes_3d, gt_labels_3d, gt_bboxes, gt_labels

Collect3D

  • add: img_meta (the keys of img_meta is specified by meta_keys)

  • remove: all other keys except for those specified by keys

Test time augmentation

MultiScaleFlipAug

  • update: scale, pcd_scale_factor, flip, flip_direction, pcd_horizontal_flip, pcd_vertical_flip with list of augmented data with these specific parameters

Extend and use custom pipelines

  1. Write a new pipeline in any file, e.g., my_pipeline.py. It takes a dict as input and return a dict.

    from mmdet.datasets import PIPELINES
    
    @PIPELINES.register_module()
    class MyTransform:
    
        def __call__(self, results):
            results['dummy'] = True
            return results
    
  2. Import the new class.

    from .my_pipeline import MyTransform
    
  3. Use it in config files.

    train_pipeline = [
        dict(
            type='LoadPointsFromFile',
            load_dim=5,
            use_dim=5,
            file_client_args=file_client_args),
        dict(
            type='LoadPointsFromMultiSweeps',
            sweeps_num=10,
            file_client_args=file_client_args),
        dict(type='LoadAnnotations3D', with_bbox_3d=True, with_label_3d=True),
        dict(
            type='GlobalRotScaleTrans',
            rot_range=[-0.3925, 0.3925],
            scale_ratio_range=[0.95, 1.05],
            translation_std=[0, 0, 0]),
        dict(type='RandomFlip3D', flip_ratio_bev_horizontal=0.5),
        dict(type='PointsRangeFilter', point_cloud_range=point_cloud_range),
        dict(type='ObjectRangeFilter', point_cloud_range=point_cloud_range),
        dict(type='ObjectNameFilter', classes=class_names),
        dict(type='MyTransform'),
        dict(type='PointShuffle'),
        dict(type='DefaultFormatBundle3D', class_names=class_names),
        dict(type='Collect3D', keys=['points', 'gt_bboxes_3d', 'gt_labels_3d'])
    ]