An effective and reliable method for fake video detection.
Train ViTHash
python train_h.py --local_rank=0 --path=../Celeb-DF --type=Celeb-DF --hash_bits=512 --pretrained=xTest ViTHash
python test_h.py --local_rank=0 --path=../Celeb-DF --type=Celeb-DF --hash_bits=512 --pretrained=xThe key of the JSON file corresponds to the name of the source video, and the value is the hash center of the source video.
Pretrained models, hash centers and datasets
Due to some software restrictions, the original sharing link is invalid and changed to Google storage.
| Item | Link | Comments |
|---|---|---|
| DeepFake | https://drive.google.com/file/d/11pPA8buSTzhzypB-D91DjJEFR4AOFVGJ/view?usp=share_link | Google drive |
| Video Splicing | https://drive.google.com/file/d/1-3mggFEiVRTPnR3PQ3yNHKsUyJ84jUkY/view?usp=share_link | Google drive |
| Similar Scenes | https://drive.google.com/file/d/1-0VBLygXSrQrYOpQqmQY8q_qv5r0uOGo/view?usp=share_link | Google drive |
| Video Inpainting | https://drive.google.com/file/d/1-35iQpD7RQTGGAFmJKSd3rptahnJuhmj/view?usp=share_link | Google drive. Extension ofDAVIS2016 |
The person with different scenes.
Different fake videos are made from the same source.
The first gif of the boat is the original video, and the remaining five videos are generated by different object inpainting methods.
The bird, person, soldier and planes are all spliced by manual.
We have developed an additional tool to compare the difference between real video and fake video. Please see tool