Depthwizard (GeoDepth-3D)

Single-view height estimation and 3D flythrough for disaster management. Built for Smart India Hackathon 2026, Problem Statement SIH26175 (ISRO) by team Doom's Dhoom.

GeoDepth-3D upgrade: primary depth model is now Depth Anything V2 (DINOv2-Large), with a mesh-generation step (Open3D/Trimesh → GLB) and a dual viewer — Three.js for relative models, CesiumJS for georeferenced, globe-accurate DSMs. A LangGraph agent layer + Bhashini (Govt. of India) voice input let you query results conversationally. See FRAMEWORK_AND_FLOW.md for the full architecture. Everything from the original deployed system (auth, projects, disaster-management services, MiDaS/im2height fallback) is kept.

One RGB satellite/aerial photo in → an elevation surface, a navigable 3D flythrough, flood simulation, crater/impact detection, building collapse-damage mapping with auto-alerts, a population estimate, a disaster classifier, and a multilingual AI assistant out.

depthwizard/
├── frontend/     Next.js + React + TypeScript + Three.js (React Three Fiber)
├── backend/      FastAPI + PostgreSQL + PyTorch (MiDaS / im2height / U-Net)
├── docker-compose.yml    runs the whole stack locally
└── .github/workflows/ci.yml   basic CI (backend tests + frontend build)

Each folder has its own detailed README (frontend/README.md, backend/README.md). This file covers the two things that span both: running everything locally, and deploying it live.

---

Run locally

Fastest: Docker Compose (needs backend/.env filled in first — copy backend/.env.example and add your keys):

docker compose up --build

• Frontend: http://localhost:3000
• Backend docs: http://localhost:8000/docs

Without Docker (two terminals):

# Terminal 1 — backend
cd backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env   # fill in SECRET_KEY, HF_API_TOKEN, OPENTOPOGRAPHY_API_KEY, etc.
uvicorn app.main:app --reload

# Terminal 2 — frontend
cd frontend
npm install
echo "NEXT_PUBLIC_API_BASE_URL=http://localhost:8000/api/v1" >> .env.local
echo "NEXT_PUBLIC_USE_BACKEND=true" >> .env.local
npm run dev

---

Deploy live — step by step

Three stops: GitHub (host the code) → Render (backend + database) → Vercel (frontend). Total time: 20-30 minutes once you have accounts on all three (GitHub, Render.com, Vercel — all free tier).

1. Push to GitHub

cd depthwizard          # this folder, containing frontend/ and backend/
git init
git add .
git commit -m "Initial commit: Depthwizard integrated system"

Then on github.com: New repository → name it depthwizard → do not initialize with a README (you already have one) → Create repository. GitHub will show you a remote URL; run:

git remote add origin https://github.com/<your-username>/depthwizard.git
git branch -M main
git push -u origin main

Your .env files are excluded by .gitignore — only .env.example goes up, so your API keys never touch GitHub. Good practice, keep it that way.

2. Deploy the backend + database on Render

1. render.com → New → PostgreSQL → name it depthwizard-db → free plan → Create Database. Once it's up, copy its Internal Database URL.
2. render.com → New → Web Service → connect your GitHub repo → set Root Directory to backend → Render will detect the Dockerfile automatically (Environment: Docker).
3. Add environment variables (Render dashboard → Environment tab) — copy every value from your local backend/.env, except DATABASE_URL, which you replace with the Internal Database URL from step 1.
4. Deploy. Once live, Render gives you a URL like https://depthwizard-backend.onrender.com. Visit https://depthwizard-backend.onrender.com/docs to confirm it's up.

3. Deploy the frontend on Vercel

1. vercel.com → Add New Project → import your GitHub repo → set Root Directory to frontend.
2. Add environment variables:
   • NEXT_PUBLIC_API_BASE_URL = https://depthwizard-backend.onrender.com/api/v1 (your real Render URL from step 2, plus /api/v1)
   • NEXT_PUBLIC_USE_BACKEND = true
3. Deploy. Vercel gives you a URL like https://depthwizard.vercel.app.

4. Connect them (the one step people forget)

The backend only accepts requests from origins listed in CORS_ORIGINS. Go back to Render → your backend's Environment tab → set:

CORS_ORIGINS=https://depthwizard.vercel.app

Redeploy the backend (Render does this automatically when you save an env var change). Now open your Vercel URL — sign up, upload an image, and it runs through the real backend end to end.

---

Environment variable checklist (for Render)

Variable	Where it comes from
SECRET_KEY	any long random string
DATABASE_URL	Render PostgreSQL → Internal Database URL
CORS_ORIGINS	your Vercel URL, set after step 3
OPENTOPOGRAPHY_API_KEY	opentopography.org → My Account
HF_API_TOKEN	huggingface.co → Settings → Access Tokens (Inference preset)
HF_CHAT_MODEL, HF_TRANSLATE_MODEL_PREFIX	defaults in .env.example are fine
ALERT_SMTP_HOST/PORT/USER/PASSWORD/FROM_EMAIL/AUTHORITY_EMAIL	Gmail App Password setup (optional — alerts degrade gracefully without it)
DEPTH_MODEL_ADAPTER	midas (working default) or im2height (needs a real trained checkpoint — see backend/app/ml/adapters/im2height_adapter.py)

Known limitations (read before demoing)

• im2height has no publicly available pretrained weights — MiDaS is the working default. Training im2height on GAMUS is a separate, multi-day task (see the chat history / backend README for the full breakdown).
• Render's free tier spins down after inactivity — the first request after idle time can take 30-60 seconds to wake up. Mention this if demoing live to judges, or ping the backend a minute before your demo.
• Every other limitation (SRTM edge cases, population estimate being a heuristic, disaster classification being rule-based) is documented in backend/README.md under "Known limitations."

Team — Doom's Dhoom

Name	Role
Axar Tandon	Project Lead / Domain Analyst
Prashant Yadav	Frontend / Presenter
Harshit Chauhan	Frontend / UI-UX Designer
Harshita Mishra	Research Analyst
Aastik Chhibar	Backend
Harsh Kumar Rai	Backend / Documentation