[general]
name=EasySegment
qgisMinimumVersion=3.44
description=Interactive HITL labeling and training plugin for geospatial segmentation using SAM3 and DINOv3-sat
version=0.1.3
author=Andras Nemes
email=contact@nemes.com.au
about=EasySegment connects QGIS to a GPU inference backend for human-in-the-loop geospatial segmentation. Use it purely as a labeling tool, or run the full pipeline.

	Workflow: create a project → define classes → draw exhaustive annotation regions → label with SAM3 or polygon tool → train → run batch inference → promote best model → correct and retrain.

	Labeling: SAM3 interactive segmentation via multi-click prompts directly on the QGIS viewport (left-click adds, right-click negates). Polygon draw tool available as an alternative. Labels are stored as polygons.

	Training: DINOv3-sat backbone pretrained on 493M satellite images with a UperNet head. Only the head trains — 50–100 labelled features per class is enough to start.

	Inference: Batch inference over any raster extent. Results load as a styled vector layer with a confidence heatmap for guided correction.

	Raster support: Works with any QGIS layer — GeoTIFF, WMS, XYZ tile URLs. Captures the viewport directly for SAM3 prompts and training chips.

	Backend: Requires a separate FastAPI + GPU server (https://github.com/anemes/backend_samdino). Supports remote hosting with optional API key authentication.
tracker=https://github.com/anemes/qgis_sam3plugin/issues
repository=https://github.com/anemes/qgis_sam3plugin
tags=segmentation,labeling,machine learning,remote sensing,SAM,SAM3,DINO,HiTL
homepage=https://github.com/anemes/qgis_sam3plugin
category=Raster
icon=icons/settings.svg
experimental=True
deprecated=False
changelog=see changelog
