How bansal-yash/terraquery-natural-language-interpretation-of-satellite-imagery works
TerraQuery is a satellite-imagery interpretation system: users submit an image and natural-language question, and the application presents an answer with supporting visual analysis. The documented VQA workflow extracts object classes, detects and segments image regions, computes geometric properties, then uses a vision-language model to answer. The repository also contains a separate GroundingDINO/SAM/Groq orchestration variant. API wiring is not visible in the excerpts, so those links are marked unverified; the README provides no further implementation detail.
Main entry points: vqa/orchestrator.py VQA orchestrator · vqa/orchestrator.py Grounded VQA variant · App.jsx Web interface · api.py HTTP backend
Components 15
User experience
- Web interface
Team_27/frontend/src/App.jsx - HTTP backend
Team_27/backend/api.py - Unified VQA API
Team_27/vqa/unified_api.py
Image question workflow
- VQA orchestrator
Team_27/vqa/orchestrator.py - Grounded VQA variant
code/Grounded-Segment-Anything/vqa/orchestrator.py
Image analysis
- Image type classifier
vqa/image_type_classifier.py - Object grounding
Team_27/vqa/orchestrator.py - Mask merging
vqa/mask_merging.py - Mask geometry
vqa/geometric_utils.py
Model integrations
- SAM3 client
Team_27/vqa/sam3_api.py - SAM3 service
Team_27/sam3/api.py - Local vision-language model
vqa/local_vlm.py - Grounded segmentation
vqa/gsam.py - Groq text service
vqa/groq_client.py
Other components
- Imagery analyst
Connections 14
- Imagery analyst → Web interface: submits image and question
Team_27/frontend/src/App.jsx - Web interface → HTTP backend: sends analysis request
- HTTP backend → Unified VQA API: routes request
- Unified VQA API → VQA orchestrator: dispatches analysis
- VQA orchestrator → Image type classifier: classifies image
Team_27/vqa/orchestrator.py - VQA orchestrator → Object grounding: finds objects
Team_27/vqa/orchestrator.py - VQA orchestrator → SAM3 client: requests masks
Team_27/vqa/orchestrator.py - SAM3 client → SAM3 service: sends segmentation request
- VQA orchestrator → Mask merging: merges masks
Team_27/vqa/orchestrator.py - VQA orchestrator → Mask geometry: computes properties
Team_27/vqa/orchestrator.py - VQA orchestrator → Local vision-language model: answers question
Team_27/vqa/orchestrator.py - Grounded VQA variant → Grounded segmentation: segments detections
code/Grounded-Segment-Anything/vqa/orchestrator.py - Grounded VQA variant → Local vision-language model: asks crop question
code/Grounded-Segment-Anything/vqa/orchestrator.py - Grounded VQA variant → Groq text service: generates and synthesizes text
code/Grounded-Segment-Anything/vqa/orchestrator.py