[general]
name=NaGeoAI
qgisMinimumVersion=3.28
qgisMaximumVersion=4.99
description=NaGeoAI is a Korean-localized QGIS plugin derived from GeoAI, providing AI-powered geospatial analysis, change detection, and first-run OpenSR setup.
version=1.9.1
author=Na Youngwoo수정
email=ds2eye@gmail.com

about=NaGeoAI is a Korean-localized customization derived from the open-source GeoAI QGIS plugin. It provides AI-powered geospatial analysis tools:

    Built-in Dependency Installer:
    - One-click installation of all AI dependencies
    - First-run checks and downloads for opensr-model and its verified checkpoint
    - 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

    Change Detection:
    - Compare co-registered two-date raster images by absolute or normalized band difference
    - Save a GeoTIFF change mask and optional GeoJSON change polygons

    Additional Analysis:
    - Text-prompt object detection, scene classification, pixel regression
    - Planetary Computer STAC download, cloud masking, temporal compositing, super-resolution

    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.

tracker=https://github.com/opengeos/geoai/issues
repository=https://github.com/opengeos/geoai

hasProcessingProvider=no

tags=nageoai, ai, deep learning, segmentation, object detection, moondream, vision language model, machine learning, remote sensing, geospatial, instance segmentation, water segmentation, deepforest, samgeo, mask rcnn

homepage=https://opengeoai.org
category=Analysis
icon=icons/icon.png

experimental=False
deprecated=False

changelog=
    1.7.0
        - Fork: rename plugin to NaGeoAI and separate the installed folder as nageoai to avoid collision with the upstream GeoAI plugin
        - Fork: add Korean localization layer (i18n.py) for menus, toolbars, and dialog titles
        - Fork: specify encoding=utf-8 and errors=replace on all subprocess calls to prevent cp949 decode failures on Korean Windows
        - Upstream: version synced with geoai-py 0.42.0; no upstream plugin code change in this release
    1.6.2
        - Fix QGIS plugin install failures on paths with spaces and NumPy version skew (#855)
        - Pin pixi QGIS to 3.44 to avoid the 3.42.2 qgz save regression (#856)
    1.6.1
        - Version-only bump; release pipeline now skips existing files on PyPI upload (#848). No plugin code change
    1.6.0
        - Resolve QGIS plugin GPU dependency install failure caused by the PyTorch wheel index (#835)
    1.5.2
        - Fix dependency installation failing on GPU installs when the PyTorch wheel index shadowed newer shared packages such as requests, blocking geoai-py/datasets resolution (#827, #829, #831, #833)
    1.5.1
        - Bump patch release for GeoAI package and QGIS plugin
    1.5.0
        - Bump minor release for GeoAI package and QGIS plugin
    1.4.4
        - Improve built-in dependency installer SSL fallback for PyPI, PyTorch, and uv certificate errors (#786)
        - Improve Windows/Python 3.12 dependency resolution diagnostics for the QGIS managed environment
        - Allow macOS QGIS dependency verification to continue when optional SAM3 import fails (#797)
    1.4.3
        - Bump patch release for GeoAI package and QGIS plugin
    1.4.2
        - Fix QGIS installer SSL trust for PyTorch wheels (#775)
    1.4.1
        - Fix QGIS stale GeoAI venv detection (#773)
        - Add inference range drawing to QGIS segmentation panels (#771)
        - Isolate diagnostics package probes from native crashes (#769)
    1.4.0
        - Add QGIS plugin diagnostics report for GitHub issue sharing
        - Redact user home paths in diagnostics output
        - Add dedicated diagnostics menu icon
        - Fix diagnostics dialog close handling after worker cleanup
    1.3.0
        - Fix QGIS plugin reload cleanup (#741)
    1.2.3
        - Fix macOS QGIS installer dependency installation
    1.2.2
        - Fix macOS About QGIS menu collision by keeping GeoAI actions out of the application menu (#723)
    1.2.1
        - Fix AttributeError on PyQt6 (#715): qualify remaining short-form Qt enums (DockWidgetArea, ScrollBarPolicy, CursorShape, MouseButton) missed in #707
    1.2.0
        - Qt5/Qt6 dual compatibility for QGIS 4.0
    1.1.0
        - Add fine-tuning support for pretrained segmentation models (#693)
    1.0.9
        - fix: run SamGeo vector export in venv subprocess (#688)
    1.0.8
        - Add support for terrainseg loss functions
        - fix: validate training paths to prevent PermissionError crash
        - fix: adaptive GPU defaults and OOM retry for DeepForest predict_tile
    1.0.7
        - Increase Moondream timeouts and add heartbeat mechanism
    1.0.6
        - Add Linux fallback to system Python when standalone download fails
    1.0.5
        - Reorder plugin menu and buttons
    1.0.4
        - Fix SamGeo vector output for GeoPackage and Shapefile formats
    1.0.3
        - Use uv for faster QGIS plugin dependency installation (#577)
    1.0.2
        - Support GEOAI_CACHE_DIR env variable for custom install location (#573)
    1.0.1
        - Fix Moondream model loading issue with transformers 5.0+ (#571)
    1.0.0
        - Add a one-click dependency installation system (#562)
        - Add support for mac mps backend (#562)
    0.7.0
        - Add instance segmentation panel to QGIS plugin (#550)
    0.6.0
        - Add Water Segmentation panel to QGIS plugin (#545)
        - Fix SAM3 load in QGIS when pkg_resources is unavailable (#530)
        - Fix plugin import issue (#546)
    0.5.0
        - Add DeepForest segmentation panel to QGIS plugin (#498)
        - Add support for pixel-level regression tasks (#493)
    0.4.0
        - Add support for exporting data in COCO and YOLO formats (#488)
        - Fix loading model progress bar issue (#488)
    0.3.1
        - Fix CUDA device detection issue
    0.3.0
        - Add smooth vector option to QGIS plugin
        - Fix point segmentation bug
    0.2.1
        - Disable SamGeo 1 and 2 due to compatibility issues
    0.2.0
        - Fix geoai import name conflict issue
        - Add support for SamGeo segmentation
    0.1.0 - Initial release
        - Moondream integration for image captioning, querying, detection, and pointing
        - Semantic segmentation model training
        - Semantic segmentation inference
