Mustatil: A New Kind of YOLO and AI Detection Program with OWLv2, Grounding DINO, and LAE-DINO Support

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 — designed for images and map areas too large for conventional computer-vision tools.
Mustatil also includes experimental support for additional AI vision models beyond standard YOLO. The Google OWL-ViT / OWLv2 model enables open-vocabulary object detection from text prompts. Grounding DINO adds powerful text-guided detection for flexible object search, while LAE-DINO provides an advanced DINO-based workflow with project-based dataset creation and training support. These models extend Mustatil from a YOLO GIS workspace into a broader AI detection and training environment.

https://doi.org/10.5281/zenodo.20481110

The installer .exe does download Python and all dependencys then it starts the GUI. Don’t worry, it takes some time and should be an option for non Python natives. Download from Github releases or on itch.

Download Windows Installer:

https://github.com/tarekwasfy01/Mustatil-YOLO-AI-Model-Trainer-/releases/download/Mustatil-5/Mustatil_5_Setup.exe

Aditionally there is a GeoPackage converter for QGis if there is a Problem with the files. Normally you can change the EPGS for a layer in QGis.
Mustatil means rectangle — a reference to both archaeological mustatils and the rectangular detection boxes used in AI object detection.
The Program was written using AI.

Thanks for sharing this — it’s great to see a geospatial workflow built around Ultralytics YOLO.

For anyone interested in the text-prompt side of this, YOLO-World supports open-vocabulary detection, and YOLOE is also worth a look for newer prompted/open-set workflows. If you test Mustatil with ultralytics + YOLO26, feel free to share a few example results here too.