Research preview
DualForensics: Deepfake Video Detection
Uniformly sample 16 frames from your clip, crop faces to 224×224, and run the DualForensics spatiotemporal model. You get a REAL / DEEPFAKE verdict, GradCAM overlays on the EfficientNet feature map, temporal self-attention weights per frame, and region-wise heatmap scores across forehead, eyes, nose, mouth, and jaw.
API:
https://parvptl-dualforensics-api.hf.spaceDrop an MP4 here, or click to browse
Max ~50 MB · 16 uniformly sampled face frames