• [2025/06/10] Initial release of model checkpoint and inference code.

UFM(UniFlowMatch) is a simple, end-to-end trained transformer model that directly regresses pixel displacement image that applies concurrently to both optical flow and wide-baseline matching tasks.

First, recursively clone this repository and install the dependencies and the UniCeption library. It is a library contains modular, config-swappable components for assembling end-to-end vision networks.

git clone --recursive https://github.com/UniFlowMatch/UFM.git
cd UFM

# In case you cloned without --recirsive:
# git submodule update --init

conda create -n ufm python=3.11 -y
conda activate ufm

# install UniCeption
cd UniCeption
pip install -e .
cd ..

# install uniflowmatch
pip install -r requirements.txt
pip install -e .

Then, verify your install by running

python uniflowmatch/models/ufm.py

Verify that ufm_output.png looks like example/example_ufm_output.png.

If you find our repository useful, please consider giving it a star ⭐ and citing our paper in your work:

@inproceedings{zhang2025ufm,
 title={UFM: A Simple Path towards Unified Dense Correspondence with Flow},
 author={Zhang, Yuchen and Keetha, Nikhil and Lyu, Chenwei and Jhamb, Bhuvan and Chen, Yutian and Qiu, Yuheng and Karhade, Jay and Jha, Shreyas and Hu, Yaoyu and Ramanan, Deva and Scherer, Sebastian and Wang, Wenshan},
 booktitle={TBD},
 year={2025}
}