GeoAI.js: Bringing Earth Observation AI to the Browser with WebGPU

We just released geoai.js, an open-source JavaScript library that brings GeoAI to the browser and Node.js, powered by Hugging Face’s 🤗 transformers.js.

It currently supports tasks like:
Image feature extraction (find similar features in satellite, aerial, or drone maps)
Object detection (cars, ships, buildings, etc.)
Solar panel and land cover detection
Change detection and segmentation

Links:
Docs: https://docs.geobase.app/geoai
Live demos: https://docs.geobase.app/geoai-live/
Repo: https://github.com/decision-labs/geoai.js

The goal is to fill the gap in the JavaScript-native GeoAI ecosystem. Would love to hear how GIS devs and remote sensing folks might use this in their workflows.

Transcript

1. GeoAI.js Bringing Earth Observation AI to the Browser with WebGPU S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
2. What is GeoAI.js? S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
3. Supported Tasks already include Object Detection Classification Segmentation and Advanced Tasks S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
4. Object Detection - Car Detection - Ship Detection - Building Detection - Oil Storage Tank Detection - Solar Panel Detection Classification - Land Cover Classification Segmentation - Image Segmentation - Wetland Segmentation - Building Footprint Segmentation - Mask Generation Advanced Tasks - Zero-shot Object Detection - Oriented Object Detection - Image Feature Extraction S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
5. The Problem • Most GeoAI solutions require complex cloud infrastructure • Python environments create barriers for web developers • Web integration of geospatial AI remains challenging • Huge community of JavaScript Developers left behind S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
6. The Solution Build on existing standards & familiar tools • Just Extend existing frontend libraries with AI capabilities like leaflet, deck.gl or maplibre. • Add tasks specific for Geospatial AI (e.g. object detection, land cover classification, etc.) • Add support for different imagery providers (e.g. ESRI, Mapbox, Google Maps) S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
7. Live Demo https://docs.geobase.app/geoai-live S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
8. How Does It Work? S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
9. 1. Model Conversion for GeoAI.js / Transformers.js • Convert to ONNX format: Supported Frameworks: TensorFlow, PyTorch, JAX • Quantization Scripts: huggingface/ transformers.js/scripts • Host Models: https:// huggingface.co/geobase S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
10. 2. Write javascript code to run the models on the imagery import { geoai } from "geoai"; const pipeline = await geoai.pipeline([{ task: "object-detection" }], { provider: "esa", }); const result = await pipeline.inference({ inputs: { polygon: yourPolygon }, postProcessingParams: { confidence: 0.8 }, mapSourceParams: { zoomLevel: 18 }, }); S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
11. 3. Display the results on a map S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
12. Key Benefits • No Cloud Infrastructure - Run AI locally in the browser • Real-time Analysis - Instant results without API calls • Cost Effective - No per-request pricing • Privacy First - Data stays in the browser • Easy Integration - Works with any mapping provider S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
13. Getting Started npm install geoai S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025
14. Thank You! Questions? • GitHub: https://github.com/geobase/geoai.js • Documentation: https://docs.geobase.app/geoai • Live Playground: https://docs.geobase.app/geoai-live • Models: https://huggingface.co/geobase/models • Email: shoaib@decision-labs.com S. Burq & M. Hassan | decision-labs.com | Big Data from Space 2025