/Group-Activity-Recognition

CVPR 2019 Paper - Learning Actor Relation Graphs for Group Activity Recognition

Primary LanguagePython

Learning Actor Relation Graphs for Group Activity Recognition

Source code for the following paper(arXiv link):

    Learning Actor Relation Graphs for Group Activity Recognition
    Jianchao Wu, Limin Wang, Li Wang, Jie Guo, Gangshan Wu
    in CVPR 2019

Dependencies

Prepare Datasets

  1. Download volleyball or collective dataset file.
  2. Unzip the dataset file into data/volleyball or data/collective.

Get Started

  1. Stage1: Fine-tune the model on single frame without using GCN.

    # volleyball dataset
    python scripts/train_volleyball_stage1.py
    
    # collective dataset
    python scripts/train_collective_stage1.py
  2. Stage2: Fix weights of the feature extraction part of network, and train the network with GCN.

    # volleyball dataset
    python scripts/train_volleyball_stage2.py
    
    # collective dataset
    python scripts/train_collective_stage2.py

    You can specify the running arguments in the python files under scripts/ directory. The meanings of arguments can be found in config.py

Citation

@inproceedings{CVPR2019_ARG,
  title = {Learning Actor Relation Graphs for Group Activity Recognition},
  author = {Jianchao Wu and Limin Wang and Li Wang and Jie Guo and Gangshan Wu},
  booktitle = {CVPR},
  year = {2019},
}