Spaces:
Sleeping
Sleeping
Commit History
-
Use face-swap-trained model and full-frame HF inference
Akshat Trivedi +1: Claude Opus 4.7 (1M context)Switches default HF model to dima806/deepfake_vs_real_image_detection and feeds the full sampled frames to the model instead of tight face crops. Tight crops were stripping the blending/edge artifacts these detectors look for, causing all-real predictions on actual deepfakes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>83c6520 -
Add backend code, deepfake_detection package, and dockerignore
Akshat Trivedi +1: Claude Opus 4.7 (1M context)Wires the FastAPI backend (src/backend/main.py) and the supporting deepfake_detection package source so the Docker build can import the model code and preprocessing pipeline. The HuggingFace fallback inference path activates when no custom checkpoint is present. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>ade59fe