How jubjones/spoton_ml works
Project type and purpose
• This is an open-source, Python-based machine-learning experimentation framework (“spoton_ml”) that provides configurable pipelines for exploratory data analysis (EDA), object detection (using Faster R-CNN), and tracking/re-identification.
• You drive experiments via small CLI scripts (e.g. run_eda.py, run_training_fasterrcnn.py), backed by a YAML-driven configuration system and integrated with PyTorch and MLflow.
File structure
• Top-level directories:
• configs/ – all YAML configs for different experiments
• scripts/ – helper scripts (e.g. exporting weights)
• src/
– core/runner.py – orchestrates pipeline execution
– data/ – data loaders & dataset definitions
– detection/ – detection strategy implementations
– evaluation/ – metric calculations for detection & tracking
– pipelines/ – EDA, detection, and tracking/reid pipeline definitions
– training/ – PyTorch training engine & runner
– utils/ – config loading, device selection, logging, MLflow hooks, reproducibility helpers
• tests/ – pytest suite for each module
• Config files (comparison_run_config.yaml, fasterrcnn_training_config.yaml, etc.) control which pipeline, parameters, and resources get used.
README
• Bare-bones README (“# spoton_ml”) but naming and directory layout imply a modular, pipeline-based ML experimentation tool.