Deepfake detector: in-browser model files
ONNX files used by the in-browser version of the twin-stream deepfake detector (source: https://github.com/DeshnaDey/BE.FR, app: https://huggingface.co/spaces/deshnadey/deepfake-detector).
| File | What | Origin and license |
|---|---|---|
face_siglip_w8.onnx |
Face stream: SigLIP deepfake classifier, 8-bit weight-only quantised (mean abs. change in P(fake) vs fp32: 0.007) | Converted from prithivMLmods/deepfake-detector-model-v1, Apache-2.0 |
scene_cf.onnx |
Scene stream: AI-generated image detector (ViT-S) | Unmodified copy of onnx/model.onnx from buildborderless/CommunityForensics-DeepfakeDet-ViT, MIT. Community Forensics (Park & Owens, CVPR 2025) |
yunet.onnx |
Face detector | YuNet 2023mar, MIT |
fusion_head_web.json |
Logistic-regression fusion head trained by the project team | Project team |
Held-out test set (357 images; separate datasets and splits from training): accuracy 81.2%, ROC-AUC 0.906 (fusion_metrics_web.json).