Paper

Spatially and Spectrally Consistent Deep Functional Maps
[Mingze Sun]¹, [Shiwei Mao]¹, [Puhua Jiang]^1,2, [Maks Ovsjanikov]³, [Ruqi Huang]^1
1 Tsinghua Shenzhen International Graduate School, China, ² Peng Cheng Laboratory, China,
³ LIX, Ecole polytechnique, IP Paris, France
ICCV, 2023

Overview

In this paper, we formulate a simple yet effective two-branch design of unsupervised DFM based on our theoretical justification, which introduces spatially cycle consistency.

Qualitative visualizations of segmentation results:

Installing Dependencies

Dependencies can be installed using:

pip install -r requirements.txt

Datasets

Within each dataset folder, the following structure is expected:

SCPAE_r/
├── shapes_train
└── shapes_test
└── corres

Training

python train.py --config scape_r

Testing

python eval.py --config scape_r --model_path ckpt.pth --save_path results_path

Citation

If you find this repository useful, please consider citing our paper:

@inproceedings{sun2023spatially,
  title={Spatially and Spectrally Consistent Deep Functional Maps},
  author={Sun, Mingze and Mao, Shiwei and Jiang, Puhua and Ovsjanikov, Maks and Huang, Ruqi},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={14497--14507},
  year={2023}
}