[Paper] [Project Page]

The project is release under the Apache 2.0 license.

Results and reference models are available in the model zoo.

Please refer to INSTALL for installation and DATA for data preparation.

Please refer to GETTING_STARTED for detailed usage.

We provide test_video.py to inference a single video. Download the checkpoints and put them to the ckpt/. and run:

python ./test_video.py ${CONFIG_FILE} ${CHECKPOINT_FILE} --video_file ${VIDOE_NAME} --label_file ${LABLE_FILE} --rendered_output ${RENDERED_NAME}

Arguments:

• --video_file: Path for demo video, default is ./demo/demo.mp4
• --label_file: The label file for pretrained model, default is demo/category.txt
• --redndered_output: The output file name. If specified, the script will render output video with label name, default is demo/demo_pred.webm.

For example, we can predict for the demo video (download here and put it under demo/.) by running:

python ./test_video.py config_files/sthv2/tsm_tpn.py ckpt/sthv2_tpn.pth

The rendered output video:

We really appreciate developers of MMAction for such wonderful codebase. We also thank Yue Zhao for the insightful discussion.

This repo is currently maintained by Ceyuan Yang (@limbo0000) and Yinghao Xu (@justimyhxu).

@inproceedings{yang2020tpn,
  title={Temporal Pyramid Network for Action Recognition},
  author={Yang, Ceyuan and Xu, Yinghao and Shi, Jianping and Dai, Bo and Zhou, Bolei},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2020},
}