High-performance multi-modal inference engine for automated OTDR cable process verification. Supports IBR, Flat-Ribbon, and Multi-Tube cable types through a unified in-memory architecture.
The latest development metrics (mAP, Top-1 accuracy) are maintained here: 👉 Model Performance & Latest Status
For detailed resource analysis and latency benchmarks, see: 👉 Pipeline Stabilization & Performance Report
- Unified In-Memory Orchestration: Eliminates disk I/O bottlenecks by processing all images in RAM.
- Stateless Validation: Real-time membership-based validation against JSON configurations.
- High Concurrency: Parallelized downstream classification using
ThreadPoolExecutor. - Production Hardened: Resolved YOLO fusion errors and implemented model warm-up logic.
- Developer First: Built on
uvfor lightning-fast environment setup and deterministic builds.
High-performance FastAPI implementation optimized for sub-second responses.
# Start the API server
uv run uvicorn api:app --host 0.0.0.0 --port 8000- Interactive Docs:
http://localhost:8000/docs - Docker Ready:
docker compose up --build
Headless batch processing tool for auditing entire cable datasets with detailed JSON summaries.
# Example: Multi-Tube Batch Processing
uv run python inference.py --cable_type multi_tube --config configs/multi_tube/config.json --save_visapi.py: FastAPI entry point with model preloading and compatibility shims.inference.py: Unified batch orchestrator for localized audits.pipelines/: Specialized logic foribr,flat_ribbon, andmulti_tube.configs/: JSON definitions for cable color codes and sequence limits.models/: Shared and cable-specific YOLOv8/v11 weights.learnings.md: Historical record of architectural wins and project milestones.
main: Full development environment (Training + API + Research).deployment: hyper-lean, production-optimized branch (API service only).
Developed by Infutrix for HFCL OTDR process verification.