Video detection and database match
Step 1 runs in the Python video worker (per-face crops). Faces are clustered with POST /api/deepface/verify. Step 2 calls POST /api/deepface/search once per unique person using the same embedding settings as Register and a built index.
1 · Detect pending→2 · Match (needs unique faces)
Video
MP4 and common formats. Size limit depends on VIDEO_SERVICE_BASE_URL.
Pick a video file to enable detection. Start the FastAPI worker on port
8787 if you have not already.Server caps MAX_DETECT_W/H. TensorFlow-backed detectors usually need Python ≤ 3.12.
Database search
Facenet · mtcnn · exact · k=5 · euclidean
Common API codes: NO_EMBEDDINGS, MISSING_INDEX, FACE_NOT_DETECTED.