Segment Anything, at the tactical edge: tap an object in field imagery and get a precise mask — computed entirely on the device with a sideloaded MobileSAM decoder.
MobileSAM brings the Segment Anything Model family to Android at a size that makes sense in the field (~40 MB decoder). Tap a vehicle, a structure, a damaged roof section — the model returns a pixel-accurate mask you can use for annotation, measurement, and analysis.
mobile_sam_decoder.tflite to activateThe GeoCam/GeoPhoto workflow hands captured imagery directly to MobileSAM: photograph a scene, segment the object of interest, and carry the annotation forward with the photo's geotag into your mission records.
During ATC-45 assessments, photograph a damaged structure and segment the specific failure — the collapsed wall section, the compromised roofline. The mask documents precisely what the assessor flagged, attached to a geotagged record in DuckDB.
Field imagery is often the most sensitive data a team collects. On-device segmentation means the photo never leaves the phone to be processed — consistent with the plugin's architecture of local models, local data, and loopback-only services.