Awesome

Detector-Free Weakly Supervised Group Activity Recognition

Dongkeun Kim, Jinsung Lee, Minsu Cho, Suha Kwak

Project Page | Paper

Overview

This work introduces a detector-free approach for weakly supervised group activity recognition.

Citation

If you find our code or paper useful, please consider citing our paper:

@InProceedings{Kim_2022_CVPR,
author    = {Kim, Dongkeun and Lee, Jinsung and Cho, Minsu and Kwak, Suha},
title     = {Detector-Free Weakly Supervised Group Activity Recognition},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month     = {June},
year      = {2022},
pages     = {20083-20093}
}

Requirements

• Ubuntu 16.04
• Python 3.8.5
• CUDA 11.0
• PyTorch 1.7.1

Conda environment installation

conda env create --file environment.yml

conda activate gar

Install additional package

sh scripts/setup.sh

Download dataset

• Volleyball dataset <br/> Download Volleyball dataset from: <br/> https://drive.google.com/file/d/1DaUE3ODT_H5mBFi8JzOVBNzVldxfbPbX/view?usp=sharing
  Dataset should be located following the file structure described below. <br/>

• NBA dataset <br/> The dataset is available upon request to the authors of "Social Adaptive Module for Weakly-supervised Group Activity Recognition (ECCV 2020)".

Download trained weights

sh scripts/download_checkpoints.sh

Run test scripts

• Volleyball dataset (Merged 6 class classification)

    sh scripts/test_volleyball_merged.sh
  

• Volleyball dataset (Original 8 class classification)

    sh scripts/test_volleyball.sh
  

• NBA dataset

    sh scripts/test_nba.sh
  

Run train scripts

• Volleyball dataset (Merged 6 class classification)

    sh scripts/train_volleyball_merged.sh
  

• Volleyball dataset (Original 8 class classification)

    sh scripts/train_volleyball.sh
  

• NBA dataset

    sh scripts/train_nba.sh
  

File structure

│── Dataset/ <br/> │ │── volleyball/ <br/> │ │ └── videos/ <br/> │ │── NBA_dataset/ <br/> │ │ └── videos/ <br/> │ │ └── train_video_ids <br/> │ │ └── test_video_ids <br/> │── checkpoints/ <br/> │── scripts/ <br/> │── dataloader/ <br/> │── models/ <br/> │── util/ <br/> train.py <br/> test.py <br/> README.md <br/> environment.yml <br/>

Acknowledgement

This work was supported by the NRF grant and the IITP grant funded by Ministry of Science and ICT, Korea (NRF-2021R1A2C3012728, NRF-2018R1A5A1060031, IITP-2020-0-00842, IITP-2021-0-00537, No. 2019-0-01906 Artificial Intelligence Graduate School Program-POSTECH).