Skip to content

Latest commit

 

History

History
129 lines (101 loc) · 6.89 KB

README.md

File metadata and controls

129 lines (101 loc) · 6.89 KB
Raw

Preparing Skeleton Dataset

@misc{duan2021revisiting,
      title={Revisiting Skeleton-based Action Recognition},
      author={Haodong Duan and Yue Zhao and Kai Chen and Dian Shao and Dahua Lin and Bo Dai},
      year={2021},
      eprint={2104.13586},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Introduction

We release the skeleton annotations used in Revisiting Skeleton-based Action Recognition. By default, we use Faster-RCNN with ResNet50 backbone for human detection and HRNet-w32 for single person pose estimation. For FineGYM, we use Ground-Truth bounding boxes for the athlete instead of detection bounding boxes.

Prepare Annotations

We provide links to the pre-processed skeleton annotations, you can directly download them and use them for training & testing.

For Kinetics400, since the skeleton annotations are large, we do not provide the direct download links on aliyun. Please use the following link to download the k400_kpfiles_2d.zip and extract it under $MMACTION2/data/skeleton/kpfiles for Kinetics400 training & testing: https://openxlab.org.cn/datasets/OpenMMLab/Kinetics400-skeleton

If you want to generate 2D skeleton annotations of specified video, please install mmdetection and mmpose first, then use the following script to extract skeleton annotations of NTURGB+D video:

python ntu_pose_extraction.py S001C001P001R001A001_rgb.avi S001C001P001R001A001.pkl

please note that, due to the upgrade of mmpose, the inference results may have slight differences from the provided skeleton annotations.

The Format of Annotations

Each pickle file corresponds to an action recognition dataset. The content of a pickle file is a dictionary with two fields: split and annotations

  1. Split: The value of the split field is a dictionary: the keys are the split names, while the values are lists of video identifiers that belong to the specific clip.
  2. Annotations: The value of the annotations field is a list of skeleton annotations, each skeleton annotation is a dictionary, containing the following fields:
    1. frame_dir (str): The identifier of the corresponding video.
    2. total_frames (int): The number of frames in this video.
    3. img_shape (tuple[int]): The shape of a video frame, a tuple with two elements, in the format of (height, width). Only required for 2D skeletons.
    4. original_shape (tuple[int]): Same as img_shape.
    5. label (int): The action label.
    6. keypoint (np.ndarray, with shape [M x T x V x C]): The keypoint annotation. M: number of persons; T: number of frames (same as total_frames); V: number of keypoints (25 for NTURGB+D 3D skeleton, 17 for CoCo, 18 for OpenPose, etc. ); C: number of dimensions for keypoint coordinates (C=2 for 2D keypoint, C=3 for 3D keypoint).
    7. keypoint_score (np.ndarray, with shape [M x T x V]): The confidence score of keypoints. Only required for 2D skeletons.

Visualization

For skeleton data visualization, you need also to prepare the RGB videos. Please refer to [visualize_heatmap_volume] for detailed process. Here we provide some visualization examples from NTU-60 and FineGYM.

Pose Estimation Results


Keypoint Heatmap Volume Visualization


Limb Heatmap Volume Visualization


Convert the NTU RGB+D raw skeleton data to our format (only applicable to GCN backbones)

Here we also provide the script for converting the NTU RGB+D raw skeleton data to our format. First, download the raw skeleton data of NTU-RGBD 60 and NTU-RGBD 120 from https://github.com/shahroudy/NTURGB-D.

For NTU-RGBD 60, preprocess data and convert the data format with

python gen_ntu_rgbd_raw.py --data-path your_raw_nturgbd60_skeleton_path --ignored-sample-path NTU_RGBD_samples_with_missing_skeletons.txt --out-folder your_nturgbd60_output_path --task ntu60

For NTU-RGBD 120, preprocess data and convert the data format with

python gen_ntu_rgbd_raw.py --data-path your_raw_nturgbd120_skeleton_path --ignored-sample-path NTU_RGBD120_samples_with_missing_skeletons.txt --out-folder your_nturgbd120_output_path --task ntu120

Convert annotations from third-party projects

We provide scripts to convert skeleton annotations from third-party projects to MMAction2 formats:

  • BABEL: babel2mma2.py

TODO:

  • FineGYM
  • NTU60_XSub
  • NTU120_XSub
  • NTU60_XView
  • NTU120_XSet
  • UCF101
  • HMDB51
  • Kinetics