VGMShield / README.md
pypy's picture
Update README.md
537ac1f verified
---
license: apache-2.0
datasets:
- OpenGVLab/InternVid
- TempoFunk/webvid-10M
pipeline_tag: video-classification
tags:
- Fake Video Detection
- Fake Video Source Tracing
---
<div style="text-align: center;">
<img src="./symbol.png" alt="symbol" style="height: 100px;"/>
</div>
# VGMShield: Mitigating Misuse of Video Generative Models
This repository pre-trained checkpoints to evaluate our detection and source tracing models. Our paper can be found at [here](https://arxiv.org/abs/2402.13126).
**Detection Model**:
[I3D](./detect/i3d/invid_i3d_i2v_i2v_best_model.pth) (0 True 1 False)
[MAE](./detect/mae/invid_mae_i2v_i2v_best_model.pth) (0 True 1 False)
[XCLIP](./detect/xclip/invid_xclip_i2v_i2v_best_model.pth) (0 True 1 False)
[MAE-sora](./detect/mae/detection_ft_sora.pt) (0 True 1 False)
**Source Tracing Model**
> 0 Hotshot-xl 1 i2vgen-xl(i2v) 2 i2vgen-xl(t2v) 3 LaVie 4 SEINE 5 Show-1 6 Stable Video Diffusion 7 VideoCrafter(i2v) 8 VideoCrafter(t2v)
[I3D](./source_tracing/i3d/invid_i3d_st_best_model.pth)-based source tracing model
[MAE](./source_tracing/mae/invid_mae_st_best_model.pth)-based source tracing model
[XCLIP](./source_tracing/xclip/invid_xclip_st_best_model.pth)-based source tracing model
[MAE](./source_tracing/mae/source_tracing_ft_sora.pt)-based(sora) source tracing model sora is label 9.