How exadel-inc/compreface works
Project type and purpose
• CompreFace is a full-stack, Docker-based face recognition system
• Provides REST APIs for detection, recognition, verification, landmark, mask, pose, age/gender
• Delivers a management UI (Angular), two Java microservices (Admin & TrainService), a Python embedding-calculator service, pluggable ML containers, and Postgres DB
• Orchestrated with docker-compose for CPU/GPU, with optional cloud deployments and load-testing/monitoring
File structure
• Top-level: docker configs, .env, docker-compose.yml, CI/CD pipelines under .github/workflows
• Custom-builds folder: multiple model plugin containers (FaceNet, InsightFace, Mobilenet, etc.) with their own Dockerfiles & docker-compose
• dev folder: dev-oriented compose files, Nginx templates, startup scripts, multi-platform configs
• db folder: Postgres Dockerfile & init scripts
• java/common, java/admin, java/api: Maven-based Java projects
– common: shared entities, DTOs, exceptions, Feign clients
– admin: Spring Boot service for user/app/role management + OAuth2
– api (TrainService): Spring Boot service exposing core face APIs, caches, async, migration, notifications
• embedding-calculator: Python Flask microservice for on-the-fly embedding computations
• ui: Angular application with Nginx Dockerfiles for dev/prod, feature-based modules, NgRx store for state management
• load-tests: K6 scenarios, Grafana dashboards for performance monitoring
README insights
• Confirms Docker-Compose architecture, CPU/GPU support, plugin-based models
• Points to docs/Architecture-and-scalability.md for high-level architecture
• Highlights REST API, role management, and plugin extensibility