YOLO / Ultralytics models
Train and run YOLO models for object detection, large images, satellite imagery, tiled inference, annotation review, and GIS export workflows. Suitable for YOLOv8, YOLOv11, and other Ultralytics-compatible models.
GIS-level AI vision workspace
Mustatil
Annotation, YOLO training, large-scale detection, satellite-map analysis, and visual pipeline building in one desktop application — designed for images and map areas too large for conventional computer-vision tools.
YOLO training
GIS detection
AI pipelines
Satellite maps
About
A desktop workspace for dataset creation, model training, geospatial inference, map-based review, and graphical AI pipelines.
Mustatil is an integrated GIS-level AI vision workspace for annotation, YOLO training, large-scale detection, satellite-map analysis, and visual pipeline building. It combines dataset creation, model training, geospatial inference, map-based review, and graphical AI pipelines in one desktop application.
Mustatil means rectangle — a reference to both archaeological mustatils and the rectangular detection boxes used in AI object detection.
Core features
Practical AI and GIS tools for very large images, satellite maps, annotations, training, and export.
- GIS-level object detection workspace
- Large-image and satellite-map detection
- Annotation workflow with classes and training data
- YOLO, RF-DETR, R-CNN, Mask R-CNN, U-Net, SAM2, ADAF and DINO workflows
- GeoPackage and GIS export workflows
- Map-based visual review and detection correction
- Graphical AI pipeline builder
- Desktop installer workflow for non-Python users
AI model support
Mustatil is built as a broad AI vision workspace: YOLO training, transformer-based detection, classical detection networks, segmentation models, prompt-based AI tools, and GIS-ready inference workflows.
RF-DETR
DETR-style object detection workflow for modern transformer-based detection. Useful as an alternative to YOLO for project-based datasets and advanced object detection experiments.
Faster R-CNN / R-CNN
Classical region-based object detection support for workflows where proposal-based detection is useful, especially as a comparison or alternative to YOLO-style detection.
Mask R-CNN
Instance-segmentation workflow for object masks, not only rectangular bounding boxes. Useful where object outlines or separated instances are more important than simple boxes.
U-Net
Semantic-segmentation workflow for pixel-level masks, raster-style classification, and image-to-mask tasks. Useful for remote sensing and map/image segmentation workflows.
SAM2
Segment Anything 2 workflow for interactive and AI-assisted segmentation. Useful for mask creation, object separation, annotation assistance, and segmentation-based review.
ADAF
Additional AI detection/filtering workflow inside the Mustatil ecosystem for advanced detection experiments, filtering, and model-assisted review.
Google OWL-ViT / OWLv2
Experimental open-vocabulary object detection from text prompts. Useful for searching objects without training a fixed class-specific model first.
Grounding DINO
Text-guided detection for flexible object search and prompt-based localization. Extends Mustatil beyond fixed-label YOLO workflows.
LAE-DINO
DINO-based detection and training workflow with project-based dataset creation, model testing, and training support.
Detect Anything / LocateAnything
Promptable detection workflow for flexible object localization. Designed as an experimental AI vision extension beside the regular YOLO/GIS workflow.
Universal AI trainer workflow
Mustatil is designed to combine annotation, class management, dataset preparation, training, detection, review, and export in one desktop workflow instead of separating these steps across many tools.
Install
Use the installer for a guided desktop setup, or install directly through Python / Conda.
Python package
py -m pip install mustatil
mustatilConda
conda install mustatil::mustatil
The Windows installer downloads Python and all dependencies, then starts the GUI. The first launch can take some time and is intended as a simple option for users who do not normally work with Python.
Downloads
Latest Mustatil 5.6 installers and project links.
Windows installer
Mustatil 5.6 setup .exe for Windows.
macOS installer
Mustatil 5.6 Apple macOS .pkg installer.
Linux installer
Mustatil 5.6 Linux .deb installer.
itch.io
Alternative download page and project presentation.
Project websites
Main Mustatil websites.
GitHub releases
Release files, installer history, and source code.
Screenshots
Mustatil workspace views for detection, annotation, training, map analysis, and AI pipelines.
Scientific archive
Permanent Zenodo DOI and citation reference.
Notes
Additional GIS compatibility and project information.
Additionally, there is a GeoPackage converter for QGIS if there is a problem with files. Normally, the EPSG code for a layer can be changed directly in QGIS.
The program was written using AI.