How luckye993/gcxl2025 works
Project type and purpose
• This is a Python-based object-detection and inference/control pipeline built around the YOLO framework. It provides:
• A real-time inference loop (camera → detect → control → communicate),
• Data-collection utilities for building a training set,
• A training/export/validation pipeline (via YOLO CLI) to produce .pt, .onnx, and OpenVINO models.
File structure
• Top-level directories:
• config/ ← JSON and YAML files for runtime parameters
• notebook/ ← Jupyter notebooks for prototyping and state-machine testing
• src/ ← main application code (split into config, core, utils, entry-point)
• tests/ ← unit tests for controller logic and main program
• Key modules under src/core:
• camera.py ← captures frames from a camera or video source
• detect.py ← loads a model (ONNX/OpenVINO) and runs inference
• control.py ← applies business logic or a state machine to detection results
• communication.py ← sends control commands to external hardware or network
• Utils under src/utils:
• dataCollect.py ← scripts to capture and label images for training
• helpers.py ← shared helper functions
• Entry-point: src/main.py reads config, wires up modules, starts the loop.
README for additional insights
• The README confirms:
• You must point config/settings.py at the correct config file.
• Training and export are done via the YOLO CLI (ultralytics).
• There are separate commands for .pt→.onnx and .pt→OpenVINO (int8 quantization).
• Validation of each model format is also supported via the CLI.