README.md

Dense Interspecies Face Embedding (DIFE)

Project Page | Paper | Poster | Presentation in Korean

An official PyTorch implementation of the paper "Dense Interspecies Face Embedding".

Dense Interspecies Face Embedding
Sejong Yang¹, Subin Jeon¹, Seonghyeon Nam², Seon Joo Kim^1
1Yonsei University, ²York University
in NeurIPS 2022

Features

• Inference for single image
• Training
• Experiment
  • Interspecies keypoint transfer
  • Interspecies face parsing
• Demo with streamlit

Setup

Docker

• Build image

$ cd docker
$ docker build -t dife .
$ cd ..

• Run docker container and access

$ docker run -ti -d --gpus=all --name=dife -v .:/workspace --ipc=host dife
$ docker exec -ti dife /bin/bash
# (Do something)

Pre-trained model

# python download.py --key model-human+dog

Test data

# python download.py --

Usage

Inference for single image

# python inference.py --image_path demo/human_000001.png
# python inference.py --image_path demo/dog_000001.png

Citation

@inproceedings{yang2022dife,
  title={Dense Interspecies Face Embedding},
  author={Sejong Yang and Subin Jeon and Seonghyeon Nam and Seon Joo Kim},
  year={2022},
  booktitle={NeurIPS},
}

Reference

• jamt9000/DVE
• facebookresearch/detectron2/projects/DensePose
• warmspringwinds/segmentation_in_style