How zhengyi-yan/plumewatch works
PlumeWatch is a Sentinel-2 coastal-plume classification and review product, with an accompanying annotation and model-development workflow. The main user flow loads or uploads a scene, classifies it with a selected model, compares and inspects map layers, then exports native-grid rasters and summaries. Separate branches prepare reviewed labels and train/evaluate models. The README documents dashboard capabilities and connections; some implementation details, especially unsampled integrations and model dispatch, are represented only where documented or marked unverified.
Main entry points: app.py Review dashboard · processing.py Class results · unet_worker.py U-Net inference · annotation_pipeline.py Mask preparation
Components 21
Scene review
- Scene input
app.py - Review dashboard
app.py - Map controls
www/app.js - Raster exports
processing.py
Classification
- Scene processing
processing.py - Random Forest
processing.py - RBF SVM
processing.py - U-Net inference
unet_worker.py - U-Net model
experiments/unet/model.py - Class results
processing.py
Annotation
- SAM candidates
integrations/cvat-sam2/auto_label.py - CVAT review
integrations/cvat-sam2/auto_label.py - Mask preparation
annotation_pipeline.py
Model development
- RF training
train_multiclass.py - SVM training
train_svm.py - U-Net training
experiments/unet/run.py - Patch sampling
experiments/unet/data.py - U-Net architecture
experiments/unet/model.py
Scene preparation
- Scene export scripts
reusable_plume_export.js
Other components
- Reviewer
- GEE exporter
Connections 24
- Reviewer → Review dashboard: uses
README.md - GEE exporter → Scene export scripts: requests scenes
README.md - Scene export scripts → Scene input: exports GeoTIFF
README.md - Reviewer → Scene input: loads scene
README.md - Review dashboard → Scene processing: calls
app.py - Review dashboard → Map controls: presents controls
README.md - Scene processing → Random Forest: classifies
processing.py - Scene processing → RBF SVM: classifies
processing.py - Scene processing → U-Net inference: dispatches inference
unet_worker.py - U-Net inference → U-Net model: runs model
unet_worker.py - Random Forest → Class results: writes predictions
processing.py - RBF SVM → Class results: writes predictions
processing.py - U-Net inference → Class results: writes rasters
unet_worker.py - Class results → Review dashboard: displays results
README.md - Review dashboard → Raster exports: downloads
app.py - Reviewer → SAM candidates: starts labeling
- SAM candidates → CVAT review: creates task
integrations/cvat-sam2/auto_label.py - CVAT review → Mask preparation: provides annotations
annotation_pipeline.py - Mask preparation → RF training: prepares labels
train_multiclass.py - Mask preparation → Patch sampling: provides masks
experiments/unet/data.py - Patch sampling → U-Net training: samples patches
experiments/unet/run.py - U-Net training → U-Net architecture: trains
experiments/unet/run.py - RF training → Class results: evaluates model
train_multiclass.py - Reviewer → SVM training: reproduces fit
README.md