How aaman-17/sih works
AI-Powered Marine Debris & Anomaly Detection is a Streamlit application for reviewing side-scan sonar and, separately, ordinary underwater photos. The main workflow takes uploaded sonar through XTF/image ingestion, preprocessing, detection and anomaly scoring, then geotags detections and exports JSON/CSV reports in the dashboard. A separate optical triage path combines generic object proposals with a plastic-film heuristic. The README documents shipwreck detection and reporting capabilities; source excerpts are partial, so some internal wiring is not shown.
Main entry points: app.py Streamlit dashboard · detection_pipeline.py Detection and fusion · geotagging.py Geotagging engine · aura_theme.py HUD styling
Components 14
Dashboard
- Streamlit dashboard
app.py - HUD styling
src/aura_theme.py
Sonar Processing
- XTF ingestion
src/xtf_loader.py - Sonar input guard
src/sonar_guard.py - Sonar preprocessing
src/preprocessing.py
Detection
- Detection and fusion
src/detection_pipeline.py - Sonar YOLO models
models - Anomaly scorer
src/autoencoder.py - Detection quality
src/detection_quality.py - Shipwreck checkpoint
models/yolo_shipwreck_ai4shipwrecks.pt
Geospatial Reporting
- Geotagging engine
src/geotagging.py - JSON/CSV reports
src/geotagging.py
Optical Triage
- Optical triage
src/optical_detector.py
Other components
- Survey operator
Connections 14
- Survey operator → Streamlit dashboard: uploads input
- Streamlit dashboard → HUD styling: injects styling
- Streamlit dashboard → XTF ingestion: loads sonar
- Streamlit dashboard → Sonar input guard: checks sonar
- Streamlit dashboard → Optical triage: runs photo triage
- Streamlit dashboard → Sonar preprocessing: preprocesses sonar
- Streamlit dashboard → Detection and fusion: runs detection
- Detection and fusion → Sonar YOLO models: detects objects
- Detection and fusion → Shipwreck checkpoint: detects shipwrecks
- Detection and fusion → Anomaly scorer: scores anomalies
- Streamlit dashboard → Detection quality: uses quality metrics
- Streamlit dashboard → Geotagging engine: geotags detections
- Geotagging engine → JSON/CSV reports: writes reports
- Streamlit dashboard → JSON/CSV reports: exports reports