How saha068-code/disaster-management-ml-project works
This repository combines a browser-based disaster damage assessment console with a Python inference server and an xBD model workflow. The principal path is an operator submitting pre- and post-disaster imagery in the console; the server preprocesses the pair, runs DamageUNet, extracts building damage regions, and returns results. The tree also includes operational console pages, event-feed endpoints, dataset acquisition, training, evaluation, and model export. Frontend behavior beyond the documented upload flow is not shown in the sampled source, so those page relationships are left unspecified.
Main entry points: serve.py HTTP server · train.py Model training · evaluate.py Model evaluation · app.js Browser console
Components 22
Assessment console
- Browser console
assets/app.js - Assessment page
app.html - Operations page
ops.html - Priorities page
priority.html - Protocol page
protocol.html - Report page
report.html - Research page
research.html - Zones page
zones.html
Inference service
- HTTP server
ml/serve.py - Pair analysis
ml/serve.py
Model workflow
- Damage U-Net
ml/model.py - Model weights
ml/outputs - Dataset loader
ml/dataset.py - Model training
ml/train.py - Model evaluation
ml/evaluate.py - Model export
ml/export.py
Data sources
- xBD dataset
ml/data/xbd - Dataset downloader
ml/download.py
Other components
- Operator
- Hugging Face xBD
- USGS events
- GDACS events
Connections 21
- Operator → Browser console: uses
- Browser console → Assessment page: loads
- Browser console → HTTP server: submits image pair
- HTTP server → Browser console: returns analysis
- HTTP server → Pair analysis: dispatches
- Pair analysis → Damage U-Net: runs inference
- Pair analysis → Model weights: loads weights
- HTTP server → USGS events: fetches events
- HTTP server → GDACS events: fetches events
- Dataset downloader → Hugging Face xBD: streams images
- Dataset downloader → xBD dataset: writes pairs
- Model training → xBD dataset: reads pairs
- Model training → Dataset loader: loads samples
- Model training → Damage U-Net: trains
- Model training → Model weights: writes checkpoints
- Model evaluation → xBD dataset: reads pairs
- Model evaluation → Dataset loader: loads samples
- Model evaluation → Damage U-Net: evaluates
- Model evaluation → Model weights: loads weights
- Model export → Damage U-Net: loads architecture
- Model export → Model weights: loads weights