Add DatasetInfo and refactor dataset base classes

[ly015]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

This PR is to merge the changes of #663 into master.

Modification

Please refer to the original PR for details.

BC-breaking (Optional)

1 DatasetInfo is needed to build dataset and inference model

DatasetInfo is defined in mmpose/datasets/dataset_info.py and is used to parse and hold dataset meta information (e.g. keypoint definition, skeleton). It is usually initialized from a config in configs/_base_/datasets/.

Please note that the old-fashioned usage listed below is still supported with warning information.

1.1 Build dataset

# New style
dataset_info = Dataset(dataset_info_cfg)
dataset = TopDownCOCODataset(
        ann_file='tests/data/coco/test_coco.json',
        img_prefix='tests/data/coco/',
        data_cfg=data_cfg_copy,
        pipeline=[],
        dataset_info=dataset_info,
        test_mode=True)

# Old style
dataset = TopDownCOCODataset(
        ann_file='tests/data/coco/test_coco.json',
        img_prefix='tests/data/coco/',
        data_cfg=data_cfg_copy,
        pipeline=[],
        test_mode=True)

Users won't have to worry about the difference in most cases, because it is encapsulated by high-level APIs like mmpose.apis.build_dataset.

1.2 Inference model

## New style
# test a single image, with a list of bboxes.
pose_results, _ = inference_top_down_pose_model(
        pose_model,
        image_name,
        person_result,
        format='xywh',
        dataset_info=dataset_info)
# show the results
vis_pose_result(
    pose_model, image_name, pose_results, dataset_info=dataset_info)

## Old style
# test a single image, with a list of bboxes.
pose_results, _ = inference_top_down_pose_model(
        pose_model,
        image_name,
        person_result,
        format='xywh',
        dataset='TopDownCOCODataset')
# show the results
vis_pose_result(
    pose_model, image_name, pose_results, dataset='TopDownCOCODataset') 

2 Use new base classes of datasets

Base classes for datasets are placed at mmpose/datasets/datasets/base. Please check the following comparison table for new and old class names.

New Base Class	Old Base Class
Kpt2dSviewRgbImgTopDownDataset	TopDownBaseDataset, HandBaseDataset, AnimalBaseDataset, FaceBaseDataset, FashionBaseDataset
Kpt2dSviewRgbImgBottomUpDataset	BottomUpBaseDataset
Kpt3dSviewKpt2dDataset	Body3DBaseDataset
Kpt3dSviewRgbImgTopDownDataset	HandBaseDataset*

* InterHand3DDataset, which was originally based on HandBaseDataset, is based Kpt3dSviewRgbImgTopDownDataset in this PR. Other hand datasets are based on Kpt2dSviewRgbImgTopDownDataset.

3 Use pose_link_color to replace pose_limb_color

# New style
img_vis_2d = imshow_keypoints(
    img,
    [kpts],
    skeleton=skeleton,
    pose_kpt_color=[(127, 127, 127)* len(kpts)],
    pose_link_color=[(127, 127, 127)* len(skeletons)])

# New style
img_vis_2d = imshow_keypoints(
    img,
    [kpts],
    skeleton=skeleton,
    pose_kpt_color=[(127, 127, 127)* len(kpts)],
    pose_limb_color=[(127, 127, 127)* len(skeletons)])

Checklist

Before PR:

• I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.
• Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.
• Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.
• New functionalities are covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, including docstring or example tutorials.

After PR:

• CLA has been signed and all committers have signed the CLA in this PR.

[innerlee]

We may add a section in the doc, describing how we support datasets via config, and how to define customized datasets