Image Feature Consensus with Deep Functional Maps

This repository is the official implementation of the paper “Zero-Shot Image Feature Consensus with Deep Functional Maps”.

Overview

Environment Set Up

We recommend the following commands to set up the environment.

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

Data Preparation

• TSS

sh tss.sh

• SPair-71k

sh spair.sh

Evaluations

To reproduce the main results, run

sh run_fmap_pascal.sh
sh run_fmap_JODS.sh
sh run_fmap_FG3DCAR.sh

TODOS

• Release SPair-71k Evaluation code.
• Release more results for TSS.

Citation

If you find this repo useful or relevant to your project and research, please kindly cite our paper:

@misc{cheng2024zeroshot,
      title={Zero-Shot Image Feature Consensus with Deep Functional Maps}, 
      author={Xinle Cheng and Congyue Deng and Adam Harley and Yixin Zhu and Leonidas Guibas},
      year={2024},
      eprint={2403.12038},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Acknowledgments

We would like to thank the following open source efforts for pushing the boundary of image feature learning and functional map optimization:

• sd-dino
• FmapLib