1.3.0 - SAM3 Integration & Pro Licensing (2025-12-28)
- NEW: SAM3 auto-segmentation ✅ WORKING (tested 2025-12-26)
- NEW: SAM3 text prompts ✅ WORKING (CLIP tokenizer fixed with runtime
patch!)
- NEW: SAM3 similar object detection ✅ WORKING (CLIP tokenizer
fixed!)
- NEW: SAM3 Pro licensing system with device limits and offline
validation
- NEW: Free tier (extent/AOI) + Pro tier (entire raster) for SAM3
text/similar modes
- NEW: Cancel button for tiled operations - abort long-running
processes anytime
- NEW: Scope selector (AOI vs full raster) with tiled processing for
large rasters
- NEW: Model selection dropdown with size variants per hardware
(GPU/CPU)
- NEW: SAM3 download flow with Hugging Face token prompt
- FIXED: CLIP tokenizer bug preventing SAM3 text/exemplar modes
(runtime patch applied)
- FIXED: Similar mode bbox handling and mask result processing
- FIXED: Threaded tiled processing stability and progress reporting
- ENHANCED: Automatic model detection and availability checks
- ENHANCED: Improved UI state updates for model changes

1.2.2 - Threading Configuration Fix
- FIXED: PyTorch threading initialization error when other plugins
pre-configure threads
- FIXED: "cannot set number of interop threads" RuntimeError on plugin
startup
- Enhanced thread configuration with graceful fallback for
pre-initialized environments
- Improved compatibility with other PyTorch-based QGIS plugins

1.2.1 - Online Tile Layer Support
- NEW: Support for online tile services (XYZ, WMS, WMTS)
- NEW: Automatic tile caching with proper georeferencing

1.2.0 - Multi-spectral UAV Support with Comprehensive Object
Detection
- NEW: Native multi-spectral UAV/satellite imagery support (5+ bands)
- NEW: Automatic NDVI calculation for vegetation detection using
NIR/Red bands
- NEW: Dual processing path - RGB for SAM2, full spectral for
vegetation analysis

1.2.2 - Threading Configuration Fix
- FIXED: PyTorch threading initialization error when other plugins
pre-configure threads
- FIXED: "cannot set number of interop threads" RuntimeError on plugin
startup
- Enhanced thread configuration with graceful fallback for
pre-initialized environments
- Improved compatibility with other PyTorch-based QGIS plugins

1.2.1 - Online Tile Layer Support
- NEW: Support for online tile services (XYZ, WMS, WMTS)
- NEW: Works with ESRI, Google Satellite, Bing Aerial
- NEW: Automatic tile caching with proper georeferencing

1.2.0 - Multi-spectral UAV Support with Comprehensive Object
Detection
- NEW: Native multi-spectral UAV/satellite imagery support (5+ bands)
- NEW: Automatic NDVI calculation for vegetation detection using
NIR/Red bands
- NEW: Enhanced batch processing with up to 100 objects for
vegetation
- NEW: Intelligent shape filtering to reject roads/tracks in
vegetation detection
- NEW: Dual processing path - RGB for SAM2, full spectral for
vegetation analysis
- NEW: Specialized detection helpers for Agriculture, Road, Vehicle,
Water, Vessels, and General classes
- NEW: Non-blocking queue system - submit multiple requests without
waiting
- NEW: Class-specific validation and detection algorithms optimized
for each object type
- FIXED: High-resolution reflectance value preservation (0-1 range)
and data type truncation
- FIXED: SAM2 tensor mismatch errors with multi-spectral input
- FIXED: Vessels detection shoreline noise with strict size validation
(>5000px rejection)
- FIXED: Oversized mask validation (rejects masks >10% of image area)
- FIXED: Crash prevention with robust processing state management
- FIXED: Shape attribute renamed from 'class_name' to 'class' for
consistency
- FIXED: Mathematical warnings in texture calculation
- ENHANCED: Vegetation detection with aspect ratio and solidity
filtering
- ENHANCED: Sequential processing with queue status feedback and
position tracking
- ENHANCED: Point and batch mode validation consistency
- IMPROVED: Comprehensive logging for debugging high-resolution
imagery issues
- IMPROVED: User experience with rapid clicking support and clear
status messages

1.1.0 - Major Bug Fixes and GUI Improvements
- FIXED: Multiple raster layer support - segmentation now works with
selected raster layer
- FIXED: Resolved "point is out of raster bounds" exception with
multiple rasters loaded (must be same CRS)
- FIXED: Panel focus management - controls properly lose focus after
use
- FIXED: Space key no longer re-triggers last clicked button
- FIXED: Added close/minimize button to control panel header
- FIXED: Font display issues on Hi-DPI screens
- FIXED: Enabled bounding box selection mode for rectangular area
prompts
- NEW: Refined GUI with improved user interface design
- NEW: Flexible panel width for better screen adaptation
- Enhanced panel layout and control positioning
- Improved keyboard event filtering and focus handling
- Better multi-raster workflow support
- Enhanced user experience with cleaner interface elements
- Improved cross-platform UI compatibility
- General stability and usability improvements

1.0.0 - Major Update with Intelligent Model Selection
- Intelligent model selection: SAM 2.1 for GPU, Ultralytics SAM2.1_B
for CPU
- Exceptional CPU performance: Sub-second segmentation on high-core
systems (24+ cores)
- Enhanced CPU performance with Ultralytics SAM2.1_B integration
(5-30x speedup)
- Automatic device detection (CUDA/Apple Silicon/CPU) with optimal
model selection
- SAM 2.1 support with improved accuracy for small objects and
occlusion handling
- Intelligent threading optimization: High-core CPUs (16+) use 75% of
cores for maximum throughput
- Multi-threaded processing for responsive UI during inference
- Point and BBox segmentation modes with adaptive processing
- 12 pre-defined classes + custom class creation with color coding
- Vehicle and Ship classes for transportation and maritime analysis
- Undo functionality for mistake correction with polygon-level
tracking
- Custom output folder selection with user-configurable paths
- Optional debug mask saving (disabled by default for performance)
- Automatic raster layer re-selection for streamlined workflow
- Professional shapefile export with detailed attributes and metadata
- Performance optimization based on zoom level and hardware
capabilities
- Comprehensive error handling and user feedback system
- Professional workflow features optimized for GIS analysis
- Updated dependencies: torch, torchvision, ultralytics,
opencv-python, rasterio, shapely, hydra-core
- Automatic model downloading: Ultralytics handles SAM2.1_B, plugin
manages SAM 2.1.
- Enhanced memory management and processing efficiency
- Cross-platform compatibility with Windows, macOS, and Linux

None