PyTorch Source code for "Learning Graph Convolutional Network for Skeleton-based Human Action Recognition by Neural Searching", AAAI2020
- python packages
- pytorch = 0.4.1
- torchvision>=0.2.1
-
Download the raw data from NTU-RGB+D and Skeleton-Kinetics. And pre-processes the data.
-
Preprocess the data with
python data_gen/ntu_gendata.py
python data_gen/kinetics-gendata.py.
-
Generate the bone data with:
python data_gen/gen_bone_data.py
-
Here, you can train the model searched by our method.
-
Configure the config file for different settings. For example, training model under the corss-view protocal:
python main.py --config ./config/nturgbd-cross-view/train_joint.yaml
-
Change the config file for corresponding dateset with its protocal.
python main.py --config ./config/nturgbd-cross-view/test_joint.yaml
- Devide the training data into trainging set and searching parts, as it is for a general NAS.
- Run the gcn_search.py with corresponding configuration.
- Here, we search on the NTU RGB+D dataset under the cross-view evaluation.
python gcn_search.py --config ./config/nturgbd-cross-view/search_joint.yaml
- Our work is inpired by Two-Stream AGCN and CEM-RL, we thank the authors for sharing their codes.
@article{peng2020learning,
title={Learning Graph Convolutional Network for Skeleton-based Human Action Recognition by Neural Searching},
author={Peng, Wei and Hong, Xiaopeng and Chen, Haoyu and Zhao, Guoying},
journal={The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI},
year={2020}
}
All materials in this repository are released under the Apache License 2.0.