Skip to content

A Few-Shot Human-Object Interaction Detection Based on Rules and Part States

License

Notifications You must be signed in to change notification settings

JamesZhutheThird/Rb-PaStaNet

Repository files navigation

Rb-PaStaNet: A Few-Shot Human-Object Interaction Detection Based on Rules and Part States

Abstract

Existing Human-Object Interaction (HOI) Detection approaches have achieved great progress on non-rare classes while rare HOI classes are still not well-detected. In this paper, we intend to apply human prior knowledge into the existing work. So we add human-labeled rules to PaStaNet and propose Rb-PaStaNet aimed at improving rare HOI classes detection. Our results show a certain improvement of the rare classes, while the non-rare classes and the overall improvement is more considerable.

Authors

Paper

IMVIP 2020 Proceedings Booklet is available on Google Drive (page.105-108). Our preview paper is available on arXiv.

Presentation

Our presentation is available on YouTube.

Code

Our source code is available on Github.

Getting Started

  1. Download dataset and pre-trained weights
chmod +x ./script/Dataset_download.sh 
./script/Dataset_download.sh 
  1. Install Python dependencies
pip install -r requirements.txt
  1. Train on HICO-DET
python tools/Train_pasta_HICO_DET.py --data 0 --init_weight 1 --train_module 1 --num_iteration 2000000 --model MODELNAME
  1. Test on HICO-DET
python tools/Test_pasta_HICO_DET.py --iteration 2000000 --model MODELNAME
  1. Generate detection
cd ./-Results/
python Generate_detection.py --model 2000000_MODELNAME
  1. Evaluation
cd ./-Results/
python Evaluate_HICO_DET.py --file Detection_2000000_MODELNAME.pkl

MODELNAME & Results

our_best (Baseline)

('Default: ', 0.2192628749969807)
('Default rare: ', 0.20547882727668224)
('Default non-rare: ', 0.22338018795239453)
('Known object: ', 0.23887022605058528)
('Known object, rare: ', 0.22431761439569864)
('Known object, non-rare: ', 0.2432171100513955)

BOOL

('Default: ', 0.22119324496184756)
('Default rare: ', 0.20543313796226934)
('Default non-rare: ', 0.22590080939029294)
('Known object: ', 0.2403750040644585)
('Known object, rare: ', 0.2242679416250538)
('Known object, non-rare: ', 0.24518620453337156)

EXPDECIMAL

('Default: ', 0.219379421568642)
('Default rare: ', 0.2057141813529523)
('Default non-rare: ', 0.22346124656813374)
('Known object: ', 0.23863934509345058)
('Known object, rare: ', 0.224585892813483)
('Known object, non-rare: ', 0.24283712954071357)

Citation

If you find this work useful, please consider citing:

@inproceedings{rbpastanet,
  title={Rb-PaStaNet: A Few-Shot Human-Object Interaction Detection Based on Rules and Part States},
  author={Shenyu Zhang, Zichen Zhu, Qingquan Bao},
  booktitle={IMVIP},
  year={2020}
}

Acknowledgment

Some of the codes are built upon TIN and PaStaNet.

We would like to express our very great appreciation to Cewu Lu and Yong-Lu Li for their constructive suggestions and guidance throughout this research work. We would also like to extend our thanks to Liang Xu for his assistance in terms of codes. Their patience and carefulness have been very much appreciated.

About

A Few-Shot Human-Object Interaction Detection Based on Rules and Part States

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages