Phases are inferred from artifacts and folder names; dates on files were not fully verified in git history for this consolidation.
Early HDF5 traversal, image extraction, and iterative tweaks to landmark logic appear in archive/phase1-image-reader-evolution/ (e.g., imgreader*.py, false-positive helpers). These scripts document how the steering landmark idea was stress-tested before consolidation into the main pipeline.
The interactive pipeline that extracts 240×320 frames, detects landmarks, applies false-positive heuristics, and writes Analysis/landmarks.txt, plot.html, and organized image folders lives in src/mbn_steering_analysis.py. The college file MBNSteeringDataAnalysisAlg.py is an older duplicate (quarantined under triage/).
- Live capture:
src/capture_gopro_gps.py(UDP stream + serial GPS → HDF5). - Camera-native GPS: Exports in
data/gopro_gps_exports/and loaderexperiments/read_gopro_gps_csv.py(from the college GoPro folder).
experiments/combined_datasets_pipeline.py (from CombinedDatasetsVersion.py) adds scipy/sklearn/statsmodels-style tooling on top of the same HDF5 assumptions—useful for exploratory work, not required for the minimal pipeline.
experiments/landmark_inspection.py— Re-open saved Plotly JSON and inspect per-landmark steering windows.experiments/extract_images_only.py— Export images from HDF5 without running the full landmark organizer.
See ../triage/README.md for duplicates and incomplete GoPro BLE experiments that were kept but not promoted to the main tree.