# Processing Snapshot

This directory contains a source snapshot of the scripts used for the released
BubbleID-Flow data products. It accompanies, but does not replace, the
canonical software repository:

https://github.com/UARK-NED3/BubbleID-Flow

The repository baseline at the time of this release snapshot was commit
`051c07dd8016cc27861509ac203b59b45b29e170`. The files under `scripts/` are the
archived processing snapshot for this dataset release. Run them from this
directory with `python -m scripts.<module_name>` so intra-script imports resolve
from the snapshot.

## Environment

`environment.yml` records the environment versions observed during release
preparation: Python 3.10.19, PyTorch 2.9.1 CPU, Detectron2 0.6, OpenCV 4.11.0,
NumPy 1.23.0, and pandas 2.3.3. The environment file is a dependency
specification, not a platform-independent binary lockfile.

## Main entry points

- `labelme_to_coco.py`: convert LabelMe annotations to ROI-aware COCO format.
- `train_detectron2.py`: fine-tune the one-class Mask R-CNN model.
- `evaluate_detectron2.py`: calculate COCO and union-mask metrics.
- `analyze_segmentation_robustness.py`: reproduce score-threshold and
  segmentation-robustness analyses.
- `audit_thermal_reduction.py`: calculate traceable thermal state summaries.
- `analyze_optical_thermal_association.py`: perform state-level
  optical-thermal screening.
- `analyze_multimodal_case.py`: prepare per-case optical, thermal, and AE
  summaries subject to the timing limits in `../timing_reconciliation.csv`.
- `create_bubbleid_flow_showcase_video.py`: regenerate the derived presentation
  videos under `../Videos`.

## Reproduction boundary

The release contains the 130 LabelMe files and checkpoint but does not include
a frozen file-level 104/26 train/holdout split manifest, original training
seed, augmentation record, or checkpoint-selection history. It therefore
supports reanalysis and new split construction, but does not independently
reconstruct the reported same-sequence evaluation exactly.