TransFlow

This repository provides code for: Transflow: Transformer as flow learner.

Installation

1. Clone the repository.
2. install dependencies.

pip install -r requirements.txt

Demo of flow estimation

1. Download trained model at [GoogleDrive] https://drive.google.com/drive/folders/1XbK0gDshbqZRirEvC9eA4OHB_crn9d0z?usp=sharing), put the demo.pth into checkpoints/ and put the mae_pvt.pth into the root/ folder.

2. Run the inference

python infer.py --keep_size

3. The flow estimations for demo_frames/ will be saved under demo_viz_output/demo_frames/

Training of flow estimation, support multi-frame

Refer to files with suffix of '_multiframe' for arbitary frame number setting.

python -u train_flow.py  --name NAME --stage 'STAGE_NAME' --validation 'VAL_NAME'

If you find the work is useful, please cite it as:

@inproceedings{lu2023transflow,
  title={Transflow: Transformer as flow learner},
  author={Lu, Yawen and Wang, Qifan and Ma, Siqi and Geng, Tong and Chen, Yingjie Victor and Chen, Huaijin and Liu, Dongfang},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={18063--18073},
  year={2023}
}