Created by Wenliang Zhao, Yongming Rao, Yansong Tang, Jie Zhou, Jiwen Lu
This repository contains PyTorch implementation for VideoABC: A Real-World Video Dataset for Abductive Visual Reasoning (TIP 2022).
VideoABC is built from the COIN dataset. Please download the videos from COIN, and place them in the data folder.
├── configs
├── data
│ ├── ...
│ ├── A129SM9S54A
│ ├── a158bYDFSwU
│ ├── A1AjNjJFGs4
│ └── ...
├── metadata
├── README.md
├── slowfast
├── tools
└── ...
The metadata folder contains the VideoABC question/choice pairs as well as the train/test split.
python tools/run_net.py --cfg configs/VideoABC/SLOWFAST_HDR.yaml
python tools/run_net.py --cfg configs/VideoABC/SLOWFAST_HDR.yaml TEST.CHECKPOINT_FILE_PATH path/to/checkpoint TEST.ENABLE True
Our code is based on SlowFast. Our dataset is based on COIN.
If you find our work useful in your research, please consider citing:
@article{zhao2022videoabc,
title={VideoABC: A Real-World Video Dataset for Abductive Visual Reasoning},
author={Zhao, Wenliang and Rao, Yongming and Tang, Yansong and Zhou, Jie and Lu, Jiwen},
journal={IEEE Transactions on Image Processing},
volume={31},
pages={6048--6061},
year={2022},
publisher={IEEE}
}