leminhbinh0209 / f3net

Unofficial implementation of ECCV20 paper "Thinking in frequency: Face forgery detection by mining frequency-aware clues"

Thinking in frequency: Face forgery detection by mining frequency-aware clues

European Conference on Computer Vision 2020

This implementation is mainly based on the desciption in the paper on top of XceptionNet. The inplementation has three models:

• MixBlock: Two-stream Collaborative Framework (@ block 7 and block 12). If you change the input image size, revise the width and height parameters in lines 49 and 50.
• FAD: Frequency-aware Decomposition
• LFS: Local Frequency Statistics

Source code is mainly referenced from this repo.
Download pre-trained XceptionNet from here.

Star if you find it useful. ⭐

f3net's Issues

Questions about the FAD implementation

Hello @Leminhbinh0209, thank you for publishing the code. I have some questions about FAD implementation. In the "FAD_Head" module，why does low_filter range from 0 to size//16? However，in the @yyk_wew's source code, it ranges from 0 to size//2.82

Experimental result

Have you trained the model? Can you get results similar to the paper?