GeoAI
GeoAI plugin for QGIS providing AI-powered geospatial analysis including tree segmentation (DeepForest), water segmentation (OmniWaterMask), Moondream vision-language model, Segment Anything (SAM1/SAM2/SAM3), semantic segmentation, and instance segmentation (Mask R-CNN).
This plugin provides AI-powered geospatial analysis tools:
Built-in Dependency Installer:
- One-click installation of all AI dependencies
- Automatic GPU detection (NVIDIA CUDA, Apple MPS)
Tree Segmentation (DeepForest):
- Detect tree crowns, birds, livestock, nests, and dead trees using pretrained models
- Run predictions on single images or large tiles with patch-based inference
Water Segmentation (OmniWaterMask):
- Automated water body detection from satellite and aerial imagery
- Output as raster or vector with optional OSM data integration
Moondream Vision-Language Model:
- Caption: Generate descriptions of geospatial imagery
- Query: Ask questions about images using natural language
- Detect: Detect and locate objects with bounding boxes
- Point: Locate specific objects with point markers
Segment Anything with SAM 3:
- Segment objects using text prompts (e.g., "tree", "building", "road")
- Segment using point prompts (foreground/background) or box prompts drawn on the map
- Process multiple points interactively or from vector files/layers
- Save results as raster (GeoTIFF) or vector (GeoPackage, Shapefile) with optional regularization
Semantic Segmentation:
- Train custom segmentation models (U-Net, DeepLabV3+, FPN, etc.)
- Run inference on raster imagery
- Export results as vector or raster data
Instance Segmentation (Mask R-CNN):
- Train Mask R-CNN models for instance-level object detection and segmentation
- Run inference on raster imagery with vectorized output
Requires the geoai-py Python package and PyTorch. CUDA or MPS support recommended for optimal performance.