Face recognition that refuses to phone home

What it does

The Open Source Face Recognition SDK by Faceplugin is a Python library for local face processing. It detects faces, extracts bounding boxes and facial landmarks, generates feature embeddings, and compares them to produce a similarity score. Everything runs on-premise on Windows or Linux, with an optional GPU but no mandatory cloud connection.

The interesting bit

In a market crowded with cloud vision APIs, this is a deliberate throwback to local-first biometrics. The API is minimal—just GetImageInfo and get_similarity—which suggests it is aimed at developers who want to drop face recognition into an app without managing model pipelines or network latency. The 0–100 similarity score and hardcoded threshold of 75 keep the logic bluntly simple.

Key highlights

• Runs fully offline; no data leaves the host machine.
• Single Python API surface: GetImageInfo for detection, landmarks, and embedding extraction, plus get_similarity for comparison.
• Supports JPG, PNG, BMP, and TIFF inputs.
• CPU-only operation is viable; GPU is optional.
• Default same-person threshold is set at 75 on a 0–100 scale.

Caveats

• The README is vague about model architecture, training data, and reproducible accuracy metrics, and the repository functions mainly as an SDK wrapper rather than a model implementation.
• macOS is not listed as a supported platform.
• Heavy promotion of commercial add-ons (liveness detection, mobile SDKs) is woven throughout the documentation.

Verdict

Worth a look if you need a quick, offline face-matching primitive for a Python app on Windows or Linux. Skip it if you require open model weights, macOS support, or detailed performance benchmarks.