[general]
name=AI Segmentation by TerraLab
description=Segment buildings, vegetation, or any object in your raster using AI, refine & export as vector polygons
qgisMinimumVersion=3.22
qgisMaximumVersion=4.99
version=2.2.1
author=Yvann and Lilien from TerraLab
email=yvann.barbot@terra-lab.ai
license=GPLv2


about=
    Point-and-click AI segmentation tool that turns raster objects into vector polygons.

    • Fully automatic: AI model and dependencies install themselves

    • Two modes: Manual runs entirely on your computer. Automatic is cloud-powered, and the imagery inside the zone you draw is sent to our detection service to be processed.

    • Works on any layer: local rasters (GeoTIFF, JPG, PNG) & online layers (WMS, XYZ tiles)

    • Precise: positive and negative click points control boundaries

    • Refinable: expand, contract, simplify, round corners, fill holes

    • Fast: Real-time segmentation & optimized for your hardware

    • Universal: Windows, macOS, Linux

    • Multilingual: English, French, Portuguese, Spanish, German, Italian, Dutch, Polish, Indonesian, Japanese, Chinese

changelog=
    2.2.1
    - Every button, hint and message is translated again in all 11 languages, including everything new in 2.2.0
    2.2.0
    - Fix a detection by hand: click it, then drag its corners or point at what the AI missed
    - Draw a polygon the AI missed, delete one with a right-click, merge two into one, or split one in two
    - Right angles squares every run now, and follows the building's real edges instead of a box
    - Manual mode gets the same shape settings as Automatic, so an object comes out the same either way
    - Every outline carries the number of points a person would draw, instead of thousands
    - Fill holes by size, and snap neighbours to a shared border, on the Keep step
    - The object library is built around what people search for, and shows the imagery behind each past run
    - Reopen a past run, restore its polygons, and go back to its zone
    - See the day your free detections come back, next to every balance
    - A cancelled run keeps every tile you already paid for
    - Detections appear while the run is still going, without slowing the map
    - Faster review and export, and a long run no longer fills your memory
    - Installation repairs broken environments, locked files and refused certificates on its own
    - Runs on QGIS 4, and on Linux builds that ship without some optional components
    2.1.10
    - Prompts naming several objects now run the first object with a tip, instead of being blocked
    - A detection that finds nothing now offers one-click example drawing, the best way to find unusual objects
    - The panel now picks up already-loaded imagery reliably
    - Smoother setup on machines with unusual disk configurations
    2.1.9
    - Fixed a rare crash when handling a network error on the newest QGIS builds
    - Clearer wording for your monthly free detections
    2.1.8
    - Compatibility fixes for the upcoming QGIS 4
    2.1.7
    - Manual export can now add polygons to an existing layer, not just a new one
    - Refining a detection in Manual mode opens with your exact review settings and recovers if interrupted
    - Fixed a stray review page appearing after a cancelled Automatic run
    - Improved stability when cancelling runs or switching modes
    - Interface polish and refreshed translations in all 11 languages
    2.1.6
    - Fixed undo being unreliable while the AI was still preparing your image in Manual mode
    - Cleaner, consistent button styling on the review page
    2.1.5
    - Reference-image detection is more accurate: your example now matches your objects at the right scale, and the tile containing it points at the real object instead of a pasted copy
    - Reference-image runs use at least two examples for far better results, with an optional example to exclude look-alikes once you have two
    - Reference-image runs start stricter, so you see fewer weak look-alikes
    - The detail level now sets itself correctly even when you describe your object in your own language
    - Editing a detection on an online basemap no longer freezes the app
    - Clearer guidance when an example is too small for the current detail level
    - Smoother review after hand-refining detections
    - More resilient installation, with clearer help if disk space runs out
    2.1.4
    - Automatic detection starts faster: the AI service begins warming up as soon as you describe what to segment
    2.1.3
    - Detections appear on the map faster during large runs
    - A cancelled or exited run keeps everything already found
    - One-click rescue actions when a run finds nothing
    2.1.2
    - Stop now cancels an in-progress detection instantly, even mid-request
    - Cleaner polygons for large objects, with holes filled during the run
    - More focused example drawing, with a streamlined set of examples per run
    2.1.1
    - Clearer feedback while the detection service is starting up, with a more reliable Cancel
    - Smarter suggestions when a description doesn't match anything found in view
    - Fixed occasional blank tiles when using online basemaps
    - Fixed detection gaps when using 5 or more example boxes
    - Segment library labels and prompt recognition improved across supported languages
    2.1.0
    - Improved Automatic detection quality and recall, especially in dense scenes
    - Default detail level is now tuned to the object you are detecting
    - Cleaner, more consistent detections and merged polygons
    - Stability and interface refinements
    2.0.1
    - Minor maintenance and stability improvements
    2.0.0
    - New Automatic mode: draw a zone, name what to find, and detect every instance in one run
    - Draw an example directly on your own imagery to guide detection (show one, find the rest)
    - Cloud-powered, no local install needed for Automatic mode
    - Live credit estimate and progress, with instant cancel
    - Review before you keep: tune confidence, simplify, expand or shrink, and fill holes with no re-run
    - Hand-refine detections in Manual mode, then finish
    - Detections deduplicated, smoothed, and saved as clean GeoPackage polygons
    1.5.1
    - Minor maintenance and stability improvements
    1.5.0
    - You can now read a version-by-version changelog right here in the Plugin Manager
    1.4.9
    - Stability and reliability improvements
    1.4.8
    - Connect with one-click browser sign-in (the manual key field is gone)
    - New "Copy connection link" button to finish signing in from another browser
    - Clearer sign-in wording and more readable Account settings text
    1.4.7
    - Fix: the AI Edit shortcut no longer overrides AI Edit's own menu entry when both plugins are installed
    1.4.6
    - More reliable dependency install: longer verification window so antivirus scans no longer trigger false timeouts
    - Install errors now show the real cause instead of a confusing header line
    1.4.5
    - Works behind authenticated corporate proxies, with guidance to enter your proxy credentials
    - Clearer messages for corrupt or unreadable rasters instead of a cryptic failure
    1.4.4
    - Export no longer fails on read-only or cloud-synced folders (OneDrive, Google Drive, network drives): the result is saved to a temporary folder and its location is shown
    1.4.3
    - Cancelling the install no longer pops up an error report
    - A half-finished install is resumed cleanly when you retry

tracker=https://github.com/TerraLabAI/QGIS_AI-Segmentation/issues
repository=https://github.com/TerraLabAI/QGIS_AI-Segmentation
homepage=https://terra-lab.ai/ai-segmentation

hasProcessingProvider=no

tags=segmentation, instance segmentation, image segmentation, ai, artificial intelligence, machine learning, deep learning, neural network, computer vision, object detection, feature extraction, classification, building detection, sam, sam2, segment anything, pytorch, raster, satellite, satellite imagery, satellite images, remote sensing, remote-sensing, aerial, orthophoto, ortofoto, drone, uav, sentinel, sentinel-2, landsat, geotiff, multispectral, buildings, vegetation, vectorization, vector, vetor, vector layer, polygon, polygons, digitizing, gpkg, geopackage, georeferencing, georreferenciamento, urban planning, agriculture, forestry, ia

category=Raster
icon=resources/icons/icon.png

experimental=False
deprecated=False
