Magic Georeferencer
Plugin ID: 4380
AI-powered automatic image georeferencing using MatchAnything deep learning model
(0) votes
Download latest
This plugin has an experimental version available!
Automatically georeference maps, aerial photos, and sketches using the MatchAnything deep learning model. Simply navigate to the approximate location and let AI find the matching features.
Note: First use will detect if necessary Python libraries are installed (see External Dependencies below). If not, you can attempt automatic installation through the Plugin GUI (it may take some time), otherwise refer to the readme for instructions. First use of the plugin will also involve downloading the Match Anything model weights from Huggingface, which may also take some time. Subsequent uses of the plugin will load much quicker.
Features:
- Zero manual GCP placement - fully automated
- Cross-modality support (maps to aerial, aerial to maps, sketches to maps)
- Progressive multi-scale refinement for improved accuracy
- GPU acceleration with automatic CPU fallback
- Confidence-based match filtering
Requirements:
- QGIS 3.22 or higher (Windows, macOS, Linux)
- Python 3.9+ (bundled with QGIS)
- 8 GB RAM minimum, 16 GB recommended
- ~2 GB disk space for model cache
- Optional: NVIDIA GPU with CUDA for faster processing
External Dependencies (auto-installed on first run):
- PyTorch >= 2.0.0 and TorchVision >= 0.15.0 (deep learning framework)
- Transformers >= 4.30.0 (HuggingFace model support)
- OpenCV >= 4.8.0 (image processing)
- SciPy >= 1.9.0 (scientific computing)
The plugin automatically downloads the AI model weights (~400 MB) from HuggingFace on first use.
Platform Support: Works on Windows, macOS, and Linux. GPU acceleration requires NVIDIA CUDA-capable GPU (optional, CPU fallback available).
Developed using Claude Code
License: Apache License 2.0
