How jolibrain/deepdetect works
Project Type and Purpose
• DeepDetect is a C++11-based, open-source machine learning server and API.
• It exposes REST/JSON and CLI interfaces for both training and inference, orchestrates data connectors, model generators, and multiple ML backends (Caffe, TensorFlow, PyTorch, XGBoost, etc.), and manages filesystem‐based models, logging, stats, and templates.
File-Structure Analysis
• Top‐Level Directories:
• clients/ – Python REST client library
• demo/ – Sample apps (imgdetect, imgsearch, segmentation, etc.)
• docker/ – Dockerfiles & scripts for container builds
• docs/ – Architecture, API, Docker, CI documentation
• examples/ – End-to-end sample configurations
• src/ – Core C++ server code:
– http/, jsonapi/, commandlineapi/ – API layers
– dto/, resources/, service_stats/ – Data types & telemetry
– backends/ – Pluggable ML library adapters
– input/output connector strategies
– generators/, graph/ – Network definition & pipeline
• templates/ – Prebuilt network configurations
• tests/, tools/, ci/ – Unit tests, helper scripts, CI pipelines
README Insights
• Emphasizes a high-level, generic API, support for many ML libraries, and JSON communication.
• Mentions key subsystems: Docker, Python/JS clients, Swagger-based docs, model templates, similarity search, various connectors.