Awesome

Introduction

This is our new work about Face Forgery Detection, which has been accepted by ECCV 2022 (oral).

UIA-ViT: Unsupervised Inconsistency-Aware Method based on Vision Transformer for Face Forgery Detection

paper link: https://www.ecva.net/papers/eccv_2022/papers_ECCV/papers/136650384.pdf

https://arxiv.org/abs/2210.12752

Environment

pytorch==1.6.0 torchvision==0.5.0 cudatoolkit==10.1 timm==0.4.12

Test

Test code: test.py

Train

Train code: train.py

Our trained model has been released: https://drive.google.com/drive/folders/1zPx4TLEfLnJDZYpSV0LhFvrMEEDzroB0?usp=sharing

Some main code about our proposed UPCL is in utils/utils.py.

Citations

Please cite the following paper in your publications if you use the python implementations:

@inproceedings{zhuang2020UIA,
  title={UIA-ViT: Unsupervised Inconsistency-Aware Method based on Vision Transformer for Face Forgery Detection},
  author={Zhuang, Wanyi and Chu, Qi and Tan, Zhentao and Liu, Qiankun and Yuan, Haojie and Miao, Changtao and Luo, Zixiang and Yu, Nenghai},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2022},
}