About DepthWizard
डेप्थविज़ार्ड · ISRO & Smart India Hackathon 2026 (PS 26175)
DepthWizard, built by team Doom's Dhoom for Smart India Hackathon 2026 (PS 26175), turns a single optical image into an elevation surface and a navigable 3D reconstruction — then goes further for disaster response: flood-level simulation, crater and impact-damage detection, building collapse-risk mapping with automatic authority alerts, WorldPop spatial demographic analysis, a rule-based disaster classifier, and a multilingual AI assistant for interpreting the results.
How this build works
This interface talks to a FastAPI backend for height estimation, DSM generation, SRTM-based calibration, and every disaster-analysis module listed above. When no backend is connected (or a given module — like SRTM calibration or the chatbot — isn't configured), the interface falls back to clearly-generated demo data rather than failing, so it stays reviewable end-to-end at any stage of setup.
Frontend stack
- Next.js (App Router)
- React + TypeScript
- Tailwind CSS
- Three.js / React Three Fiber / Drei
- Recharts
- Lucide React
Backend stack
- FastAPI + Pydantic
- PostgreSQL + SQLAlchemy
- Rasterio / GDAL
- MiDaS + im2height (pluggable depth adapters)
- U-Net + ResNet18 segmentation
- Hugging Face Inference API (chatbot + translation)
Known limitations
- Relative height (rDSM) output does not carry a real-world coordinate reference; only a metric DSM with SRTM or ground-control-point calibration does.
- A .tif or .tiff extension does not guarantee valid GeoTIFF metadata — that is confirmed and calibrated by the backend.
- Population estimates are heuristic approximations, not census figures. Disaster classification is a transparent, rule-based signal, not a trained model — each label ships with its evidence.
- Validation metrics are only populated once a reference DSM/LiDAR raster is supplied; otherwise they show as not available rather than an invented number.