---
license: mit
language:
- en
pipeline_tag: visual-question-answering
metrics:
- accuracy
- bleu
- rouge
base_model:
- meta-llama/Llama-2-7b-hf
---

## Model details

**Model type:** M2F2-Det is a deepfake (i.e., forged facial image) detection model, which produces both fake score and explanations. The model is trained on 
[FaceForensics++](https://github.com/ondyari/FaceForensics) and [Deepfake Detection VQA](https://arxiv.org/pdf/2402.00126).

**Model date:** This preliminary version has been trained in January 2025. More updates can be expected. 

**Project page:** https://github.com/CHELSEA234/M2F2_Det

**Paper:** [Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector](https://huggingface.co/papers/2410.02761) CVPR2025 **Oral** presentation.


**Where to send questions or comments about the model:**

Please feel free to contact Xiao Guo (xiao07guo@gmail.com) with any questions or to post an issue on our code base.
[https://github.com/zhipeixu/FakeShield/issues](https://github.com/zhipeixu/FakeShield/issues)