How ranjeet-2/antariksh-jepa works
Antariksh-JEPA is a cross-modal satellite-image retrieval app: a user supplies an optical or SAR query, the app prepares and loads held-out imagery and matching model/index artifacts, then presents ranked matches and spatial context. The sampled runtime is concentrated in Web_app/app.py; Web_app/asset_cache.py implements test-tile extraction. The README documents broader training and evaluation methodology, but those workflows live mainly in notebooks whose implementation was not sampled, so they are described only as documented capabilities rather than linked into the runtime flow.
Main entry points: app.py Streamlit dashboard · asset_cache.py Test tile cache
Components 15
Query experience
- Streamlit dashboard
Web_app/app.py - Image query
Web_app/app.py
Data and assets
- Test imagery
Web_app/app.py - Dataset repository
- Test tile cache
Web_app/asset_cache.py - Model run assets
Web_app/app.py
Retrieval engine
- Contrastive checkpoint
Web_app/app.py - Antariksh-JEPA model
Web_app/app.py - Binary FAISS indexes
Web_app/app.py - Gallery metadata
Web_app/app.py
Results and mapping
- Ranked matches
Web_app/app.py - Image previews
Web_app/app.py - Esri map tiles
- Spatial results view
Web_app/app.py
Other components
- Analyst
Connections 18
- Analyst → Streamlit dashboard: submits query
README.md - Streamlit dashboard → Image query: accepts image
Web_app/app.py - Streamlit dashboard → Dataset repository: downloads data
README.md - Streamlit dashboard → Test tile cache: extracts tiles
Web_app/app.py - Test tile cache → Test imagery: writes tiles
Web_app/asset_cache.py - Streamlit dashboard → Test imagery: loads imagery
Web_app/app.py - Streamlit dashboard → Model run assets: loads bundle
Web_app/app.py - Model run assets → Contrastive checkpoint: includes checkpoint
Web_app/app.py - Model run assets → Binary FAISS indexes: includes indexes
Web_app/app.py - Model run assets → Gallery metadata: includes metadata
Web_app/app.py - Streamlit dashboard → Antariksh-JEPA model: instantiates model
Web_app/app.py - Contrastive checkpoint → Antariksh-JEPA model: loads weights
Web_app/app.py - Streamlit dashboard → Binary FAISS indexes: loads indexes
Web_app/app.py - Streamlit dashboard → Gallery metadata: loads metadata
Web_app/app.py - Streamlit dashboard → Ranked matches: shows top matches
README.md - Streamlit dashboard → Image previews: renders imagery
Web_app/app.py - Streamlit dashboard → Spatial results view: shows locations
Web_app/app.py - Spatial results view → Esri map tiles: requests tiles
Web_app/app.py