suny-sht / shic

Official implementation of the 2024 ECCV paper SHIC: Shape-Image Correspondences with no Keypoint Annotation

SHIC: Shape-Image Correspondences with no Keypoint Supervision

ECCV 2024

Authors

• Aleksandar (Suny) Shtedritski
• Christian Rupprecht
• Andrea Vedaldi

Contact: {suny, chrisr, vedaldi}@robots.ox.ac.uk

Website

For more detailed information about SHIC, visit our project website.

Demo

A demo of SHIC is available on Hugging Face. You can try it out here.

Code

Installation

conda create -n shicenv python=3.9
conda activate shicenv
pip install torch==1.12.0 torchvision==0.13.0 einops spharapy==1.1.2 trimesh==3.23.5 matplotlib numpy scipy Pillow jupyter

Weights

Download pretrained the weights here here.

Unzip and put in this folder. Should look like

.
├── ...
├── models                 
│   ├── shapes          # obj shape files
│   └── weights         # pth weights for shape features and image models
├── README.md
└── ...

Demo notebook

Run the shic.ipynb notekbook for a more custom demo of the model. The notebook runs very quickly on a CPU, which is used by default

Training code

Coming soon!

BibTeX

@inproceedings{shtedritski2024SHIC,
      title={SHIC: Shape-Image Correspondences with no Keypoint Supervision}, 
      author={Shtedritski, Aleksandar and Rupprecht, Christian and Vedaldi, Andrea},
      year={2024},
      booktitle={ECCV},
}