GeoOSAM - Advanced Segmentation for QGIS

🛰️ State-of-the-art image segmentation using Meta's SAM 2.1, SAM3, and Ultralytics models with intelligent hardware optimization

🌟 Features

🤖 SAM3 Integration: Text prompts, similar object detection, auto-segment
🔍 Similar from Selection: Use any segmented polygon as exemplar to find matching objects (v1.4)
🗺️ Vector Extent ROI: Restrict entire-raster processing to a polygon area of interest (v1.4, Pro)
🔑 Pro Licensing: Free tier (extent) + Pro tier (entire raster)
⚙️ Model Size Selection: Choose from 4-5 sizes for both Meta & Ultralytics models
🗺️ Tiled Processing: Process entire rasters with auto-tiling for large imagery
📥 Automatic Downloads: All models download automatically with progress bars
🔐 Hugging Face Integration: SAM3 downloads via token authentication
🚀 Exceptional CPU Performance: Sub-second segmentation on high-core CPUs (24+ cores)
🧠 Intelligent Hardware Detection: Automatically selects best models for GPU/CPU
🛰️ Multi-spectral Support: Native 5+ band UAV/satellite imagery with NDVI calculation
🎯 Multiple Modes: Point-click, bounding box, batch, auto-segment, text prompts, similar objects
📋 12 Pre-defined Classes: Buildings, Roads, Vegetation, Water, Vehicle, Vessels, and more
🌿 Enhanced Vegetation Detection: Spectral analysis for superior vegetation mapping
🌐 Online Map Support: Works with ESRI, Google Satellite, and XYZ/WMS/WMTS tile services
↶ Undo Support: Mistake correction with polygon-level undo on both Segment and Detect tabs
📁 Custom Output: User-selectable output folders
🎨 Class Management: Custom classes with color coding
📡 Smart Workflow: Auto-raster selection, progress tracking
💾 Professional Export: Multi-format export (GeoPackage, Shapefile, GeoJSON, FlatGeobuf)
🔧 Adaptive Processing: Optimized based on zoom level and hardware

📊 Performance & Model Selection

Available Models

Hardware Type	Available Models	User Selection
GPU (>3GB)	SAM2.1 Tiny/Small/Base+/Large (Meta) + SAM3	5 models via dropdown
Apple Silicon	SAM2.1 Tiny/Small/Base+/Large (Meta) + SAM3	5 models via dropdown
CPU / Low GPU	SAM2.1_T/B/L (Ultralytics, CPU-optimized)	3 models via dropdown

Model Sizes & Performance

Model	Size	Best For	Speed (GPU)	Speed (CPU 24+)
SAM2.1 Tiny	156MB	Fast processing	0.2-0.4s	1-2s
SAM2.1 Small	184MB	Balanced	0.3-0.6s	1.5-3s
SAM2.1 Base+	323MB	High accuracy	0.5-1s	3-5s
SAM2.1 Large	898MB	Best quality	0.8-1.5s	5-8s
SAM3	~160MB	Auto-segment/Text	0.3-0.7s	N/A
SAM2.1_T (CPU)	~40MB	Fast CPU	N/A	<1s
SAM2.1_B (CPU)	~80MB	Balanced CPU	N/A	1-2s
SAM2.1_L (CPU)	~224MB	Quality CPU	N/A	2-3s

🎯 Intelligent Selection:

GPU Systems (>3GB): Dropdown offers 4 SAM2.1 sizes + SAM3 for flexibility
CPU Systems: Dropdown offers 3 CPU-optimized SAM2.1 variants (T/B/L)
Automatic Detection: Plugin detects your hardware on startup
One-Click Download: Select any model, it downloads automatically with progress bar
Default Choice: SAM3 for GPU (if available), SAM2.1_B for CPU

🚀 Quick Start

1. Install Plugin

Enable in QGIS Plugin Manager or download from GitHub

2. Load Imagery & Select Class

Supported Data Sources:

🗂️ Local raster files (GeoTIFF, JP2, etc.)
🌐 ESRI services
🗺️ Google Satellite, Bing Aerial (XYZ tiles)
🌍 WMS/WMTS tile services

3. Point & Click to Segment

Select a class (Buildings, Water, etc.) and click on objects in your imagery. Works identically with local rasters and online tile services.

4. View Results

5. Export Results

Export segmented polygons in GeoPackage, Shapefile, GeoJSON, or FlatGeobuf format

🤖 SAM3 Features

SAM3 integration with semantic segmentation

Text Prompts 📝

Status: CLIP tokenizer bug FIXED

Fix: Custom CLIP tokenizer monkey-patch applied at runtime
Functionality: Enter text like "building", "car", "tree" to find matching objects
Accuracy: High-quality semantic understanding via CLIP
Tested: ✅ Fully functional with all text prompts

How to use:

Select SAM3 from model dropdown
Enter text prompt (e.g., "buildings", "vehicles")
Choose scope: "Visible Extent (AOI)" or "Entire Raster"*
Click "Auto-Segment All Objects"

Similar Objects 🔍

Status: Exemplar mode

Fix: Same CLIP tokenizer fix enables exemplar-based search
Functionality: Click one object to find all similar objects
Technology: Uses SAM3 exemplar mode with CLIP embeddings
Tested: ✅ Fully functional

How to use:

Select SAM3 from model dropdown
Click "Find Similar" mode button
Choose scope: "Visible Extent (AOI)" or "Entire Raster"*
Click on a reference object to find all similar ones

*Pro Licensing:

GeoOSAM is free and open-source, but SAM3 Pro features help sustain development:

Free Tier (Forever):
- All SAM2/SAM2.1 features - unlimited
- SAM3 point, bbox, text prompts and similar mode on Visible Extent (AOI) - unlimited
- Perfect for most use cases!
Pro Tier - Early Access ($15):
- SAM3 text prompts and similar mode on Entire Raster with auto-tiling
- Ideal for production workflows and large-scale mapping
- ⚠️ Experimental: SAM3 is under active development - bugs expected!
- Your support directly funds bug fixes and new features
- One-time payment, lifetime v1.x updates, 1 device per license
- Works offline (30-day cache), best-effort support
- Early supporter pricing - price may increase as features stabilize
Get License: Email [email protected]
Activation: Click "Manage License" button in plugin

Test Results (v1.3):

✅ SAM3 initialization: SUCCESS
✅ Auto-segmentation: SUCCESS
✅ Text prompts (CLIP): SUCCESS (tokenizer fixed)
✅ Exemplar mode (CLIP): SUCCESS (tokenizer fixed)
✅ Entire raster processing: SUCCESS (tiled processing with 1024px tiles)

📥 SAM3 Download:

Downloads via Hugging Face with access token
Token prompt appears in QGIS on first use
Requires Read token from https://huggingface.co/settings/tokens

⚠️ Requirements:

GPU >3GB VRAM:
- ✅ NVIDIA GPU (CUDA) - Recommended, best performance
- ✅ Apple Silicon (M1/M2/M3) - Supported, good performance
- ❌ AMD GPU - Not officially supported (ROCm is unreliable in practice)
- ❌ CPU-only - SAM3 not available (use SAM2.1 instead)
Ultralytics >= 8.3.237
Hugging Face account with access token

Note: If you have AMD GPU or CPU-only, SAM3 features won't work. Use SAM2.1 with point/bbox/auto segmentation instead.

🛠 Known Issues

For current limitations and upcoming fixes, see:

👉 Known Issues

⚠️ Batch Mode Development Status

Batch mode is currently in active development with varying performance across different object classes:

🎯 Tested Classes:

Vegetation ✅ - Advanced NDVI analysis, excellent results with multi-spectral imagery
Vessels ✅ - Optimized for water body detection, good performance
Buildings/Residential 🔧 - Basic functionality, mixed results

🔧 Classes Under Development:

Other classes 🚧 - Limited testing, performance may vary

📝 Best Practice:

Point-click mode remains the most reliable for all classes
Batch mode works best on clear, high-contrast imagery
Mixed workflow recommended: use batch where it works well, point-click for precision

📋 System Requirements

Minimum

QGIS 3.16+
Python 3.7+
8GB RAM
2GB disk space

🧪 Running Tests

python3 -m pytest tests/ -v

Requires: pytest, numpy, opencv-python, rasterio, shapely

📦 Installation

⚠️ Important: Both installation methods require manual dependency installation

Option 1: From QGIS Plugin Repository (Recommended)

Open QGIS
Go to Plugins > Manage and Install Plugins
Search "GeoOSAM"
Click Install Plugin
Install dependencies (see below)

Option 2: Download from GitHub (Manual)

Download ZIP from: https://github.com/espressouk/GeoOSAM

Extract the plugin:

# Extract and rename to remove -main suffix
unzip GeoOSAM-main.zip
mv GeoOSAM-main geoOSAM
cd geoOSAM

Copy to QGIS plugins directory:

# Linux
cp -r . ~/.local/share/QGIS/QGIS3/profiles/default/python/plugins/geo_osam

# macOS
cp -r . ~/Library/Application\ Support/QGIS/QGIS3/profiles/default/python/plugins/geo_osam

# Windows
xcopy . "C:\Users\%USERNAME%\AppData\Roaming\QGIS\QGIS3\profiles\default\python\plugins\geo_osam" /E /I

Install dependencies (see below)

Required Dependencies (Both Options)

🎯 Windows: Use OSGeo4W Shell (Recommended)

IMPORTANT: Choose CPU or CUDA version based on your hardware!

# Open OSGeo4W Shell (Start Menu → OSGeo4W → OSGeo4W Shell)
# Check if you have NVIDIA GPU:
nvidia-smi

# If nvidia-smi shows your GPU, install CUDA version:
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip install "ultralytics>=8.3.237" opencv-python rasterio shapely hydra-core iopath pillow numpy

# If no NVIDIA GPU or nvidia-smi fails, install CPU version:
# pip install torch torchvision "ultralytics>=8.3.237" opencv-python rasterio shapely hydra-core iopath pillow numpy

Required for SAM3 text prompts and similar-object detection

SAM3 concept-prompt modes (text prompts, Find Similar) require the Ultralytics CLIP fork. The standard pip install clip package will not work and will cause a TypeError: 'SimpleTokenizer' object is not callable crash.

pip install git+https://github.com/ultralytics/CLIP.git ftfy wcwidth

GeoOSAM includes a runtime compatibility shim (sam3_clip_fix.py) that handles tokenizer differences across CLIP versions, but the Ultralytics fork is still required for SAM3 text features to function.

Hugging Face Access (SAM3 Weights)

To download SAM3 weights from within QGIS, create a Hugging Face Access Token with Token type: Read and paste it into the download dialog.

🍎 macOS: Use Terminal

pip3 install torch torchvision "ultralytics>=8.3.237" opencv-python rasterio shapely hydra-core iopath pillow numpy

🐧 Linux: Use Terminal

# For NVIDIA GPU with CUDA:
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu118
pip3 install "ultralytics>=8.3.237" opencv-python rasterio shapely hydra-core iopath pillow numpy

# For CPU-only:
# pip3 install torch torchvision "ultralytics>=8.3.237" opencv-python rasterio shapely hydra-core iopath pillow numpy

🔧 Alternative: QGIS Python Console (All Platforms)

# Open QGIS > Plugins > Python Console, paste and run:
import subprocess
import sys
packages = ["torch", "torchvision", "ultralytics>=8.3.237", "opencv-python", "rasterio", "shapely", "hydra-core", "iopath", "pillow", "numpy"]
for pkg in packages: subprocess.check_call([sys.executable, "-m", "pip", "install", pkg]); print(f"✅ Installed {pkg}")

⚠️ IMPORTANT: Verify CUDA for Windows/Linux GPU Users

# After installation, verify CUDA is working (QGIS Python Console):
import torch
print(f"CUDA available: {torch.cuda.is_available()}")
if torch.cuda.is_available():
    print(f"GPU: {torch.cuda.get_device_name(0)}")

Expected: CUDA available: True + your GPU name

If False: You installed CPU-only PyTorch by mistake! See Installation Guide for fix.

Model Download (Automatic)

All models download automatically with progress bars on first use.

🔄 How It Works:

Select Model Size: Choose from dropdown (4 GPU models or 3 CPU models)
Automatic Download: First-time use triggers download with progress bar
One-Time Setup: Models cached locally for instant future use
No Manual Steps: Everything happens automatically!

📥 Download Details:

Model Type	How It Downloads	Where Stored
SAM2.1 (GPU)	Direct download with progress dialog	Plugin checkpoints folder
SAM2.1 T/B/L (CPU)	Ultralytics auto-download (notification)	`~/.ultralytics/weights/`
SAM3	Hugging Face with token prompt	Plugin folder or Ultralytics cache

⚡ Performance Highlights:

Multiple Size Options: Choose speed vs accuracy tradeoff
24+ Core CPUs: Sub-second segmentation with SAM2.1_T
Intelligent Threading: Automatically uses 75% of available cores
Memory Optimized: Efficient processing for large imagery
GPU Acceleration: Full CUDA/Apple Silicon support

🔧 Manual Download (if needed):

# SAM2.1 models (GPU) are auto-downloaded on first use
# SAM2.1 T/B/L (CPU) are auto-downloaded by Ultralytics
# SAM3 requires Hugging Face token (prompted in UI)

# Manual fallback for SAM2.1 (rare):
cd ~/.local/share/QGIS/QGIS3/profiles/default/python/plugins/geo_osam/sam2/checkpoints/
wget https://dl.fbaipublicfiles.com/segment_anything_2/092824/sam2.1_hiera_tiny.pt
# Replace 'tiny' with 'small', 'base_plus', or 'large' as needed

🎯 Use Cases

🏙️ Urban Planning: Building and infrastructure mapping
🌱 Environmental Monitoring: Vegetation and land cover analysis with NDVI
🛰️ UAV/Drone Mapping: Multi-spectral imagery analysis and processing
🚗 Transportation: Vehicle and traffic analysis
🌊 Coastal Studies: Ship detection and water body mapping
🏗️ Construction: Site monitoring and progress tracking
📡 Remote Sensing: Large-scale multi-spectral imagery analysis
🌾 Agriculture: Crop monitoring with spectral vegetation indices

🛰️ Multi-spectral UAV/Satellite Support

Advanced Spectral Analysis

GeoOSAM now provides native support for high-resolution multi-spectral imagery:

Feature	Capability	Benefit
5+ Band Support	Automatic NDVI calculation	Superior vegetation detection
Reflectance Values	0-1 range preservation	Accurate spectral analysis
High Resolution	UAV imagery (0.08m/pixel)	Fine-scale object detection
Batch Processing	Up to 100 objects per selection	Efficient large-area mapping
Shape Filtering	Road/track rejection	Clean vegetation results

Supported Image Types

Multi-spectral UAV: 5+ band imagery (Blue, Green, Red, NIR, RedEdge)
Satellite Imagery: Landsat, Sentinel, Planet, etc.
Reflectance Data: Automatically handles 0-1 reflectance values
High-Resolution: Optimized for 0.05-0.1m pixel size UAV imagery
Standard RGB: Backward compatible with 3-band imagery

Intelligent Band Processing

🔹 5+ Bands: Automatic NDVI calculation using NIR/Red bands
🔹 3-4 Bands: Enhanced green channel processing
🔹 RGB: Standard texture analysis
🔹 Single Band: Grayscale texture detection

Vegetation Detection Excellence

For vegetation mapping, GeoOSAM automatically:

Calculates NDVI from NIR and Red bands
Filters linear features (roads, tracks) with shape analysis
Processes up to 100 objects in batch mode
Validates object geometry (aspect ratio, solidity)
Preserves spectral fidelity throughout processing

⚙️ Technical Details

Model Architecture

SAM 2.1 (Meta): Latest from Meta AI with 4 size variants (Tiny/Small/Base+/Large)
- Sept 2024 release with improved accuracy for small objects
- Optimized for GPU inference (CUDA/Apple Silicon)
- Sizes: 156MB - 898MB with speed/accuracy tradeoffs
SAM2.1 T/B/L (Ultralytics): CPU-optimized variants
- T=Tiny (40MB), B=Base (80MB), L=Large (224MB)
- Enhanced multi-core CPU performance
- Efficient threading for sub-second segmentation
SAM3 (Ultralytics): Advanced instance segmentation
- ✅ Auto-segmentation - WORKING (tested 2025-12-26, 5 objects detected)
- ✅ Text prompts - WORKING (CLIP tokenizer fixed in v1.3 with runtime patch)
- ✅ Similar objects mode - WORKING (CLIP tokenizer fixed in v1.3)
- Requires Ultralytics >= 8.3.237 and GPU >3GB
Automatic Selection: Based on hardware detection and user preference

Performance Optimization

Intelligent Threading: High-core CPUs (16+) use 75% of cores for optimal performance
Model Size Selection: User can choose speed vs accuracy tradeoff per task
Adaptive Crop Sizes: Zoom-level aware processing
Memory Management: Efficient handling of large imagery
Device Detection: Automatic CUDA/MPS/CPU optimization with core-count awareness
Tiled Processing: 1024x1024 tiles with 128px overlap for large rasters
Threaded Workers: Non-blocking UI during intensive operations

📚 Documentation

🤝 Contributing

We welcome contributions! Please see our Contributing Guide.

📞 Support

Issues: GitHub Issues
Email: [email protected]
Documentation: Wiki

🙏 Acknowledgments

Meta AI: For the Segment Anything Model (SAM 2.1)
Ultralytics: For SAM2.1_B integration and optimization
QGIS Community: For the excellent GIS platform
PyTorch Team: For the deep learning framework

☕ Support GeoOSAM

If you find this plugin useful, please consider buying me a coffee to support continued development and new features.

📄 License

This project is licensed under the GPL v2 License - see the LICENSE file for details.

🏆 Citation

If you use GeoOSAM in your research, please cite:

@software{geoosam2025,
  title={GeoOSAM: Advanced Segmentation for QGIS with Intelligent Model Selection},
  author={Butbega, Ofer},
  year={2025},
  url={https://github.com/espressouk/GeoOSAM}
}

🔄 Changelog

v1.4.3 - Vegetation & Bbox Improvements (2026-06-12)

✨ IMPROVED: Vegetation segmentation on multispectral imagery (5+ bands) — less shadow and soil bleed in tree and crop masks (automatic; no effect on RGB or non-vegetation classes)
🔧 FIXED: BBox selections on high-resolution imagery were sometimes rejected — small boxes now register correctly
🔧 FIXED: Batch mode could freeze the request queue after an error — now recovers cleanly

v1.4.2 - Pro Features & Fixes (2026-06-04)

🆕 NEW: Fill Holes in Filters tab (Pro) — removes interior voids from segmented polygons
🔧 FIXED: SAM3 text prompts now use native concept-prompting — text was previously ignored
🔧 FIXED: Removed debug print statements from production code
🆕 NEW: Automated test suite (41 tests)
📝 NOTE: Pro users must connect online once after updating to v1.4.2

v1.4.1 - Code Quality Fix (2026-05-09)

🔧 FIXED: Flake8 E231/E221/E226 whitespace issues in geo_osam_dialog.py and vegetation_helper.py

v1.4.0 - UI Redesign & Workflow Improvements (2026-05-09)

🎨 NEW: Redesigned 4-tab flat UI — Segment, Detect, Results, Filters, Settings as gear overlay
🔍 NEW: Similar from Selection — use any segmented polygon as exemplar
🗺️ NEW: Vector Extent ROI for Entire Raster (Pro) — restrict tiled processing to a polygon
↶ NEW: Undo and Cancel buttons on Detect tab
🔀 NEW: Find Similar toggleable — click again to exit similar mode

v1.3.6 - Code Quality & Security Fixes (2026-04-22)

🔒 FIXED: Flake8 code quality issues cleared (unused imports, bare excepts, ambiguous names)
🐛 FIXED: Status label color styles now correctly applied (broken f-string in stylesheet)

v1.3.5 - Security Fix (2026-04-22)

🔒 FIXED: Removed eval from checkpoint download script (flagged by QGIS plugin repository security scanner)
🔒 FIXED: torch.load now uses weights_only=True for safer checkpoint loading
🔒 FIXED: Suppressed Bandit B310 false positives on urllib calls using hardcoded https:// URLs
🔒 FIXED: Removed unused HuggingFace Hub download functions from build_sam.py (dead code, triggered critical security flag)
🔒 FIXED: Excluded dev-only shell scripts and .bat files from plugin package

v1.3.4 - Settings Tab, Size & Shape Filters, GSD Support (2026-04-06)

🗂️ NEW: Tabbed UI — "Segmentation" and "Settings" tabs for a cleaner, less cluttered layout
📐 NEW: GSD / Resolution card — auto-detect from raster metadata or enter manually (cm/px)
📏 NEW: Real-world size filtering — filter objects by min/max diameter (cm) or area (m²); live pixel-equivalent hints update as you type
🔷 NEW: Shape filters — min circularity, min compactness, max aspect ratio; applied after segmentation across all modes
📊 NEW: area_m2, diameter_cm, circularity attributes exported per feature
📊 NEW: Per-mask progress bar during text/similar mode processing
🔧 IMPROVED: Font sizes increased across the UI for laptop-screen readability
🔧 IMPROVED: Tooltips use HTML with explicit white background (QGIS compatible)
🔧 IMPROVED: Active filter summary shown inline in the status label

v1.3.3 - Multi-Point, Export Formats & Fixes (2026-02-07)

🎯 NEW: Multi-point segmentation — Shift+click to add positive points, Ctrl+click for negative, normal click to run prediction
🎯 NEW: Visual +/- markers on map showing accumulated points before prediction
📄 NEW: Multi-format export support (GeoPackage, Shapefile, GeoJSON, FlatGeobuf)
📄 NEW: GeoPackage as default export format
🔧 FIXED: License activation connectivity issue
🔧 FIXED: SAM3 multi-point predictions returning separate objects instead of combined result
🔧 FIXED: "NoneType has no attribute write" crash on Windows when sys.stderr is None during QGIS startup

v1.3.2 - Bug Fix (2026-02-06)

🔧 FIXED: MD5 hash security flag (added usedforsecurity=False)

v1.3.1 - Bug Fix (2026-02-06)

🔧 FIXED: Exemplar mask normalization in SAM3 similar object detection

v1.3.0 - SAM3 & Model Management Updates (2025-12-22)

🎯 Major Features:

🤖 SAM3 Support: Automatic instance segmentation (tested 2025-12-26)
- Auto-segment: ✅ PRODUCTION READY - Automatically find all objects (5 objects detected in test)
- Text prompts: ✅ WORKING - CLIP tokenizer fixed in v1.3 with runtime patch
- Similar objects: ✅ WORKING - CLIP tokenizer fixed in v1.3 with runtime patch
- Free tier (extent) + Pro tier (entire raster) licensing
⚙️ Model Size Selection: Choose optimal model for your needs
- GPU: 4 SAM2.1 sizes (Tiny/Small/Base+/Large, 156MB-898MB) + SAM3
- CPU: 3 SAM2.1 sizes (T/B/L, 40MB-224MB, Ultralytics optimized)
- User-friendly dropdown with size and performance info
🗺️ Scope Selector: Process visible area (AOI) or entire raster
- AOI mode: Fast processing of visible extent
- Full raster mode: Auto-tiling with 1024px tiles, 128px overlap
- Progress tracking for tiled operations
- 🛑 Cancel button: Abort long-running tiled operations at any time
📥 Automatic Downloads: All models download on-demand
- SAM2.1 (GPU): Progress bar with percentage complete
- SAM2.1 T/B/L (CPU): Ultralytics auto-download with notification
- SAM3: Hugging Face token prompt for secure download
- Manual download URLs provided if auto-download fails

🔧 Improvements:

✨ ENHANCED: Automatic hardware detection and model availability checks
✨ ENHANCED: UI dynamically shows/hides SAM3 features based on availability
✅ FIXED: Similar mode bbox handling and mask result processing
✅ FIXED: Threaded tiled processing for responsive UI during large raster operations
🚀 OPTIMIZED: All SAM2.1 model URLs verified and updated to September 2024 release

v1.2.2 - Threading Configuration Fix (2025-09-20)

🔧 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

v1.2.1 - Online Tile Layer Support (2025-07-20)

🌐 NEW: Support for online tile services (XYZ, WMS, WMTS)
🗺️ NEW: Works with ESRI, Google Satellite, Bing Aerial
⚡ NEW: Automatic tile caching with proper georeferencing

v1.2.0 - Multi-spectral UAV Support (2025-07-09)

🛰️ NEW: Native multi-spectral UAV/satellite imagery support (5+ bands)
🌿 NEW: Automatic NDVI calculation for vegetation detection using NIR/Red bands
🔧 FIXED: High-resolution reflectance value preservation (0-1 range)
🔧 FIXED: Data type truncation issues with multi-spectral imagery
🚀 NEW: Enhanced batch processing with up to 100 objects for vegetation
🎯 NEW: Intelligent shape filtering to reject roads/tracks in vegetation detection
🔧 FIXED: SAM2 tensor mismatch errors with multi-spectral input
🔧 FIXED: Oversized mask validation (rejects masks >10% of image area)
🔧 FIXED: Mathematical warnings in texture calculation
🌿 ENHANCED: Vegetation detection with aspect ratio and solidity filtering
📊 NEW: Comprehensive logging for debugging high-resolution imagery issues
🛰️ NEW: Dual processing path - RGB for SAM2, full spectral for vegetation analysis

v1.1.0 - Latest (2025-07-03)

FIXED: Multiple raster layer support - segmentation now works with selected raster (same CRS)
FIXED: Panel focus management - controls properly lose focus after use
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
General stability and usability improvements

v1.0.0 - Major Update with Intelligent Model Selection

Intelligent Model Selection: Automatic SAM 2.1 vs SAM2.1_B selection
Enhanced CPU Performance: SAM2.1_B integration for 5-10x CPU speedup
Ultralytics Integration: Professional computer vision library support
Improved Device Detection: Better GPU/CPU/Apple Silicon handling
Updated Dependencies: Modern ML stack with automatic model downloads

⚙️ Environment Options

To force CPU-only mode, set this environment variable before launching QGIS:

export GEOOSAM_FORCE_CPU=1

📋 Reporting Issues

Please check:

Plugin version (latest preferred)
QGIS version (3.16+ required)
Dependencies installed:
- torch
- torchvision
- ultralytics
- opencv-python
- rasterio
- shapely
- hydra-core
- iopath

Report issues at: GitHub Issues

🚀 Planned Features

To be determined based on user feedback and usage patterns.

💡 Performance Tips

Zoom wisely: Try different zoom levels to get the best results for your selected class.
Force CPU mode: If GPU memory is limited

Last updated: 2026-06-12 Plugin Version: 1.4.3 QGIS Compatibility: 3.16+

Name		Name	Last commit message	Last commit date
Latest commit History 124 Commits
.github		.github
docs		docs
help		help
helpers		helpers
i18n		i18n
sam2		sam2
screenshots		screenshots
test		test
tests		tests
.gitattributes		.gitattributes
.gitignore		.gitignore
CHANGELOG_v1.3.md		CHANGELOG_v1.3.md
CHANGELOG_v1.4.md		CHANGELOG_v1.4.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
Makefile		Makefile
README.html		README.html
README.md		README.md
README.txt		README.txt
__init__.py		__init__.py
benchmark_joss.py		benchmark_joss.py
geo_osam.py		geo_osam.py
geo_osam_dialog.py		geo_osam_dialog.py
geo_osam_dialog_base.ui		geo_osam_dialog_base.ui
geo_osam_license.py		geo_osam_license.py
icon.png		icon.png
icon_32X32.png		icon_32X32.png
icon_64X64.png		icon_64X64.png
icon_base.png		icon_base.png
metadata.txt		metadata.txt
paper.bib		paper.bib
paper.md		paper.md
pb_tool.cfg		pb_tool.cfg
plugin_upload.py		plugin_upload.py
pylintrc		pylintrc
requirements.txt		requirements.txt
resources.py		resources.py
resources.qrc		resources.qrc
resources_rc.py		resources_rc.py
sam3_clip_fix.py		sam3_clip_fix.py
setup.cfg		setup.cfg

Folders and files

Latest commit

History

Repository files navigation