Here we have something new for FF++ dataset

[johannwyh]

Hello, every developer of FaceForensics++!

First of all, thank you for your enduring effort in maintaining FF++ dataset. It has been widely recognized as one of the most critical benchmarks by the forgery detection community.

Recently, our team proposed a new N-to-N face swap algorithm, with several fresh features. Upon our own quantitative testing on 10K FF++ images (which is a common setting in recent face swap research), our algorithm achieves the top in several metrics, including ID retrieval, face shape retention, and forgery detection.

Our paper is currently under review. Inspired by your recent collaboration with FaceShifter team, we also would like to provide our generation results on FF++ videos, to help enrich your dataset and promote the continuous and rapid development of the forgery detection community. Here we provide two videos in FF++ dataset for first view, the face pasted at the left-down corner is the source identity.

• 312_021.mp4 , which is also the demo video at FaceShifter's FF++ page
• 700_813.mp4 , one selected from ten *00_*.mp4 video

Finally, we want to consult that, what materials should be provided/prepared in order to help you review our method? Please feel free to contact me under this issue or e-mail me wang_yuhan@zju.edu.cn privately for further cooperation.

Thank you very much! Wish everything well with you all!