Self-hosted face recognition that keeps biometrics off the cloud

What it does

CompreFace is a self-hosted, Docker-based face recognition server. It exposes a REST API for recognition, verification, detection, and several plugins—age, gender, mask, head pose, and landmark detection—using FaceNet and InsightFace models under the hood. A web UI adds role-based access management, letting multiple users or applications share the same backend safely.

The interesting bit

This is essentially high-quality glue code: instead of training new models, it wraps proven computer vision libraries in a scalable service. The project handles the dull but critical work—CPU and GPU model variants, client SDKs for JavaScript, Python, and .NET, and a permission system—so you don’t have to wire raw model inference to your app server.

Key highlights

• Runs fully self-hosted under Apache 2.0, keeping face data in-house
• Supports both CPU and GPU inference with configurable model backends
• Bundles plugins for mask detection, head pose, age, and gender estimation
• Includes a web panel for managing users and access roles
• Provides client SDKs for JavaScript, Python, and .NET

Caveats

• Requires a modern x86 processor with AVX support; ARM chips and older hardware need not apply

Verdict

Worth a look if you need face recognition in a private network or customer deployment and would rather integrate an API than babysit Python dependencies. Skip it if you are training custom models from scratch or running on ARM/embedded edge devices.