Atlaso Lens
The AI copilot for planet-scale imagery
Detect, monitor and query anything on Earth — across optical, radar, aerial and drone imagery — from a single AI-native workspace and API.
Sources
- Optical satellite30 cm – 10 m
- SAR radarAll-weather, day & night
- Aerial & drone1 – 10 cm
- ElevationDSM / DTM / LiDAR
- Your dataVectors, tables, labels
Atlaso Lens engineAI
- Geospatial foundation modelPre-trained on 4B km² of imagery
- Task models40+ detectors & segmenters
- Change engineSeasonal & sensor-aware baselines
- Geo-reasoningNatural-language questions
Analysis-ready data cube · STAC catalog
Outcomes
- Workspace & mapsExplore, compare, share
- Alerts & webhooksEmail, Slack, Teams, HTTP
- API & SDKsREST, Python, TypeScript
- Exports & BIGeoParquet, COG, OGC
Detect & segment
Find every object that matters, at any scale
Run production-grade detection and segmentation across a city block or a continent. Results are georeferenced, versioned and ready to analyze.
- 40+ pre-trained models
- Fine-tune with 30 labels
- Confidence on every feature
- Sub-meter to 10 m imagery
Monitor change
Continuous monitoring with less noise
Set up a monitor once and get notified only when something meaningful changes. Our change engine understands seasons, shadows, clouds and sensor differences.
- Optical + SAR fusion
- Seasonal baselines
- Per-object change history
- Alerts via email, Slack or webhook
Ask
Ask questions in plain language
Lens plans and runs the geospatial analysis for you, then answers with numbers, maps and the evidence behind them — so anyone on your team can get answers without writing a query.
- Grounded in your imagery
- Cites scenes & models used
- Answers as maps, charts or tables
- Save questions as monitors
Integrate
Insights where your team already works
Every result is a standard geospatial asset with a stable ID. Pull it via API, stream it into your warehouse or open it in your favourite GIS.
- REST API & typed SDKs
- Webhooks for every event
- OGC-compliant tile services
- Warehouse-friendly exports
POST /v1/ask → 200 OK
{
"answer": "214 new structures inside the AOI since 2025-03-01.",
"confidence": 0.96,
"evidence": {
"model": "buildings-v4",
"scenes": ["optical_20250304", "optical_20260918"],
"features": "https://api.atlaso.co.in/v1/results/res_93kd/features.geojson"
},
"usage": { "km2": 38.4, "credits": 12 }
}Model catalog
Production-ready models, with honest metrics
A selection of our most-used models. Each is evaluated on geographically diverse, held-out test sets — and you can benchmark on your own data before you buy.
| Model | Task | Sensors | Resolution | Benchmark |
|---|---|---|---|---|
| Buildings v4 | Segmentation | Optical | 0.3–1 m | 95.1% F1 |
| Vessels v3 | Detection | Optical + SAR | 0.5–10 m | 97.4% precision |
| Vehicles v4 | Detection | Optical | 0.3–0.5 m | 93.8% F1 |
| Solar panels v3 | Segmentation | Optical | 0.3–10 m | 94.6% IoU |
| Land cover (12 classes) | Segmentation | Optical + SAR | 10 m | 89.2% accuracy |
| Crop type v5 | Classification | Optical + SAR time series | 10 m | 92.0% accuracy |
| Forest loss v2 | Change | Optical + SAR | 10 m | 91.5% F1 |
| Flood extent | Segmentation | SAR | 10 m | 93.3% IoU |
| Damage assessment | Classification | Optical | 0.3–0.5 m | 88.7% accuracy |
| Vegetation height | Regression | Optical + elevation | 1 m | ±1.8 m MAE |
Need something else? Ask about custom models.
Works with open standards your stack already speaks
- GeoJSON
- GeoParquet
- Cloud-Optimized GeoTIFF
- STAC
- WMS / WMTS
- Vector tiles
- CSV
- Shapefile
What imagery does Atlaso Lens use?
We work with a network of open and commercial providers covering optical (30 cm to 10 m), SAR, aerial and drone imagery. You can also bring your own imagery via cloud storage or upload. We handle ordering, harmonization and cloud masking for you.
How accurate are the models?
Every model in the catalog ships with published evaluation metrics on held-out, geographically diverse test sets. You can also run an evaluation on your own labeled data before going to production, and track performance over time.
Can I train models for my own objects?
Yes. Because our models are built on a geospatial foundation model, you can fine-tune a new detector or segmenter with as few as 30 labeled examples using the built-in labeling tool — no ML expertise required.
How does natural-language search work?
Lens translates your question into a geospatial query plan — which areas, time ranges, models and statistics to use — runs it, and returns an answer with links to the exact scenes and features used as evidence.
Does it work with our existing GIS tools?
Results are standard geospatial formats and services, so they open directly in desktop GIS, web maps, notebooks and data warehouses. Webhooks and the API let you push insights into any internal system.
See your area of interest through a new lens
Book a 30-minute walkthrough on your own geography. We’ll run live detections and show what’s changed — no GIS expertise required.