Upload satellite images get pixel-level cloud segmentation masks and coverage percentages in under a second, powered by a fine-tuned U-Net + EfficientNet-B0 deep learning model trained on GOES-18 Band 13 infrared data.
Getting Started · Features · Architecture · API Reference · Contributing
AtmosGen takes satellite imagery as input and produces:
- Cloud Segmentation Mask — a binary mask identifying cloud vs. clear-sky pixels
- Cloud Coverage Percentage — the fraction of the image covered by clouds
- Overlay Visualization — the original image with detected clouds highlighted
All predictions are stored per-user, enabling historical tracking and trend analysis through an analytics dashboard.
- Deep Learning Inference — U-Net architecture with EfficientNet-B0 encoder; sub-second predictions on GPU
- Drag-and-Drop Upload — Upload satellite images via the browser; supports PNG, JPG, and TIFF formats
- Analytics Dashboard — Track average cloud coverage, total forecasts, processing times, and view coverage trends with interactive charts
- Forecast History — Searchable, sortable log of every prediction with detailed results
- JWT Authentication — Secure user accounts with cookie-based session management
- Dark / Light Theme — System-aware with manual toggle; dark mode default
- Responsive Design — Glassmorphism UI with Framer Motion animations, fully mobile-friendly
- FAQ & Contact — Built-in FAQ page and contact form that opens a pre-filled email
| Cloud Coverage Prediction | Analytics Dashboard |
 |
 |
AtmosGen/
├── backend/ # FastAPI server
│ ├── main.py # Routes, middleware, lifecycle
│ ├── cloud_model.py # U-Net model loading & inference
│ ├── auth_service.py # JWT token creation & validation
│ ├── mongodb_client.py # MongoDB CRUD operations
│ ├── database.py # Database abstraction layer
│ ├── schemas.py # Pydantic request/response models
│ ├── satellite_service.py # GOES-18 satellite data integration
│ └── requirements.txt
├── frontend/ # React 19 + TypeScript
│ ├── src/
│ │ ├── app/
│ │ │ ├── pages/ # Landing, Forecast, Dashboard, History,
│ │ │ │ # Login, Register, Contact, FAQ, 404
│ │ │ ├── components/ # Navigation, ThemeProvider
│ │ │ └── layouts/ # AppLayout wrapper
│ │ └── lib/ # API client, auth context
│ └── package.json
├── checkpoints/ # Trained model weights (.pth)
├── core_model/ # Training scripts & model definitions
└── data/ # Training & evaluation datasets
Satellite Image ── FastAPI ── U-Net Model ── Cloud Mask + Coverage %
│ │
└── MongoDB ──────────────────┘
│
Dashboard ── Charts & History
| Architecture | U-Net with EfficientNet-B0 encoder (segmentation_models_pytorch) |
| Pre-training | ImageNet weights on encoder |
| Fine-tuning Data | GOES-18 Band 13 (10.3 µm longwave IR) satellite imagery |
| Task | Binary cloud segmentation |
| Input | Single-channel or RGB satellite image (auto-resized to 256×256) |
| Output | Binary mask + cloud coverage percentage + overlay visualization |
| Framework | PyTorch 2.x |
| Inference | ~70ms on GPU, ~200ms on CPU |
The model performs pixel-level classification, distinguishing cloud from clear-sky regions. A probability threshold of 0.5 is applied to the sigmoid output to generate the binary mask. Cloud coverage is calculated as the ratio of cloud pixels to total pixels.
| Requirement | Version |
|---|---|
| Python | 3.10+ |
| Node.js | 18+ |
| MongoDB | Local instance or MongoDB Atlas |
| GPU (optional) | CUDA-compatible for faster inference |
git clone https://github.com/Rishabh1925/AtmosGen.git
cd AtmosGenpython -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txtCreate backend/.env:
MONGODB_URI=mongodb+srv://<user>:<pass>@<cluster>.mongodb.net/atmosgen
JWT_SECRET=your-secret-keyStart the server:
cd backend
python main.py
# API available at http://localhost:8000cd frontend
npm install
npm run dev
# App available at http://localhost:5173All endpoints are served from the FastAPI backend at http://localhost:8000.
| Method | Endpoint | Description |
|---|---|---|
POST |
/auth/register |
Create a new user account |
POST |
/auth/login |
Authenticate and receive JWT |
POST |
/auth/logout |
Clear session cookie |
GET |
/auth/me |
Get current authenticated user |
| Method | Endpoint | Description |
|---|---|---|
POST |
/predict |
Upload satellite image(s) returns cloud mask, coverage %, and overlay |
| Method | Endpoint | Description |
|---|---|---|
GET |
/dashboard |
Aggregated stats: avg coverage, total forecasts, avg processing time |
GET |
/forecasts |
List all forecasts for the authenticated user |
GET |
/forecasts/{id} |
Get a specific forecast by ID |
| Method | Endpoint | Description |
|---|---|---|
GET |
/satellite/regions |
List available GOES-18 satellite regions |
GET |
/satellite/layers |
List available satellite data layers |
- FastAPI — async Python web framework
- PyTorch + segmentation_models_pytorch — model architecture and inference
- MongoDB — document store for users, forecasts, and analytics
- JWT — stateless authentication with HTTP-only cookies
- React 19 + TypeScript — component-based UI
- Tailwind CSS — utility-first styling
- Recharts — bar charts and gradient area charts
- Framer Motion — page transitions and micro-animations
- React Router — client-side routing
- GOES-18 Band 13 — 10.3 µm longwave infrared satellite data
- U-Net + EfficientNet-B0 — encoder-decoder segmentation architecture
- PIL / NumPy — image preprocessing and postprocessing
Contributions are welcome! Here's how:
- Fork the repository
- Create a feature branch:
git checkout -b feature/your-feature - Commit your changes:
git commit -m "Add your feature" - Push to the branch:
git push origin feature/your-feature - Open a Pull Request
For bug reports or feature requests, open an issue.
This project is licensed under the MIT License — see the LICENSE file for details.
Built by Rishabh