FSFM-3C Models (Pre-trained/Fine-tuned Vision Transformers)

A self-supervised pre-training framework to learn a transferable facial representation that boosts various downstream face security tasks.

[Paper FSFM: A Generalizable Face Security Foundation Model via Self-Supervised Facial Representation Learning]

Project

Github Rrepository

Environment

Git clone our repository, creating a python environment and activate it via the following command:

conda create -n fsfm3c python=3.9
conda activate fsfm3c
pip install -r requirements.txt

Model Loading

import models_vit
from huggingface_hub import hf_hub_download
CKPT_SAVE_PATH = [your checkpoint storage path]
CKPT_NAME = [checkpoint name]
hf_hub_download(local_dir=CKPT_SAVE_PATH,
                repo_id='Wolowolo/fsfm-3c',
                filename=CKPT_NAME)
model = models_vit.__dict__['vit_base_patch16'](
    num_classes=2,
    drop_path_rate=0.1,
    global_pool=True,
)
checkpoint = torch.load(os.path.join(CKPT_SAVE_PATH, CKPT_NAME), map_location='cpu')
model.load_state_dict(checkpoint['model'])

Downstream Use

Face security tasks such as deepfake detection, face anti-spoofing, and diffusion facial forgery detection.

BibTeX entry and citation info

@article{wang2024fsfm,
  title={FSFM: A Generalizable Face Security Foundation Model via Self-Supervised Facial Representation Learning},
  author={Wang, Gaojian and Lin, Feng and Wu, Tong and Liu, Zhenguang and Ba, Zhongjie and Ren, Kui},
  journal={arXiv preprint arXiv:2412.12032},
  year={2024}
}