Extending Segment Anything Model (SAM) for GeoAI
Extending SAM for extracting features of interest from geospatial imagery
Combining SAM with pretrained ArcGIS Object Detection models
SAM can be prompted using bounding boxes to produce segmentation masks. We have combined several pretrained ArcGIS object detection models with SAM, so they can extract the polygonal boundary of features instead of bounding boxes.
Turning Object Detection models into Object Segmentation models with SAM
Text Prompt support for SAM (Grounded SAM)
The Grounded SAM model combines SAM with Grounding DINO, an open set object detection model, so it can be prompted using free form text prompts to extract features of various kinds.
Prompting SAM with Text Prompts to extract features
Here are a few examples of features extracted using text prompts:
Fine tuning SAM for Geospatial AI tasks
The SAMLoRA model in arcgis.learn applies the low-rank-adaptation (LoRA) technique to the SAM model and fine-tunes it for geospatial imagery segmentation.
Segmenting Geospatial imagery using low rank adaptation of SAM model
Building footprint extraction using fine-tuned SAMLoRA model
