Draft:Mustatil
This draft appears to contain text generated by a large language model (such as ChatGPT). You cannot use LLMs to generate, draft, or rewrite article content.
Please delete the portions which were LLM-generated, and summarize in your own words a range of independent, reliable, published sources that discuss the subject. If you have used an LLM but you believe that your use is covered by one of the two exceptions (basic copyediting and translation between two languages in which you are fully fluent), please provide a transcript of the prompt(s) you used while writing your draft (see the how-to guide). Reviewers can make mistakes. If you have not received any LLM assistance, you may seek a second opinion at the AfC Help Desk. See the advice page on large language models for more information.
Where to get help
How to improve a draft
You can also browse Wikipedia:Featured articles and Wikipedia:Good articles to find examples of Wikipedia's best writing on topics similar to your proposed article. Improving your odds of a speedy review To improve your odds of a faster review, tag your draft with relevant WikiProject tags using the button below. This will let reviewers know a new draft has been submitted in their area of interest. For instance, if you wrote about a female astronomer, you would want to add the Biography, Astronomy, and Women scientists tags. Editor resources
|  |
Mustatil is a desktop geospatial artificial-intelligence and computer-vision workspace developed by Tarek Wasfy. It is designed for annotation, model training, large-raster object detection, satellite and aerial-image analysis, human review of detections, and export to geographic information system (GIS) formats. The project was originally motivated by the remote-sensing search for archaeological mustatils in Arabia, but its workflow can be used with user-defined classes and other types of visually identifiable objects.[1]
The name has a double meaning. Mustatil (Arabic: مستطيل) means "rectangle" and is also the archaeological term used for the elongated stone monuments that motivated the project; object-detection systems commonly represent candidate objects with rectangular bounding boxes.[1]
The software combines tools that would otherwise commonly be split between an annotation application, a machine-learning environment and GIS software. Its central workflow is to annotate examples, train or load a model, analyse large images in tiles, review the resulting detections and export accepted results with geographic coordinates.[2][3]
Important: descriptions of features in this article are based mainly on project-maintained sources, package registries and the project's Zenodo software archive. They describe the software and its documented capabilities; they should not be read as independent validation of model accuracy or archaeological interpretations.
Overview
[edit]Mustatil is intended for geospatial computer-vision projects in which source imagery can be too large to process conveniently as a single image. The project supports ordinary images as well as georeferenced raster data such as GeoTIFF and BigTIFF. Large rasters can be divided into overlapping tiles for inference, after which detections are transformed back from tile coordinates into the coordinate-reference system of the source raster.[1]
The project describes its architecture as local-first. This means that imagery, coordinates and model execution can remain on the user's own computer rather than requiring the source data to be uploaded to a hosted inference service. Some functions can nevertheless require internet access, for example software installation, dependency installation, model downloads or web-map use.[1]
A central design principle is human review. Automated detections are treated as candidate locations rather than automatically accepted findings. Users can inspect, filter, edit or reject detections before exporting them to a GIS dataset.[1]
Main functions
[edit]Mustatil includes or documents workflows for:
- image and geospatial annotation;
- class management and dataset creation;
- YOLO model training and inference;
- large-image, GeoTIFF and BigTIFF tiled detection;
- satellite and aerial-image analysis;
- map-based review and correction of detections;
- video detection and annotation workflows;
- semantic and instance segmentation;
- prompt-based and open-vocabulary detection;
- graphical AI pipelines and model chaining;
- confidence, class and rule-based filtering;
- export to GeoPackage, GeoJSON, CSV, images, masks and training datasets.[2][3]
The large-raster workflow exposes tile overlap and confidence settings. Overlapping tiles can reduce missed detections at tile boundaries, while also increasing computation and potentially creating duplicate detections that must be filtered or merged.[1]
Supported AI model families
[edit]The project documentation describes support for several computer-vision model families. Availability can differ between editions and releases.
| Model or family | Documented use in Mustatil |
|---|---|
| YOLO / Ultralytics | Object detection, training, tiled inference, class filtering, review and GIS export. |
| Faster R-CNN | Region-proposal object detection, including large-image tiled inference and training workflows. |
| RF-DETR | Transformer-based object detection and model-training workflows. |
| Mask R-CNN | Instance segmentation where object masks are required in addition to bounding boxes. |
| U-Net | Pixel-level semantic segmentation for raster classification and image-to-mask workflows. |
| SAM / SAM2 | AI-assisted segmentation, including segmentation from existing boxes or annotations. |
| Grounding DINO | Text-guided object localization and prompt-based detection. |
| OWL-ViT / OWLv2 | Experimental open-vocabulary detection from text prompts. |
| LAE-DINO | Project-based DINO detection, experimentation and training workflows. |
The source repository also contains additional experimental components and plugins. Consequently, a feature present in the source tree or in Mustatil 5.6 is not necessarily present in the Microsoft Store edition of Mustatil 6.[3]
For example, the July 2026 Mustatil 6 release notes state that the new release introduced a new C++ graphical interface, multi-GPU support, faster start-up, an included standard DINO model and an all-in-one package, while also stating that ADAF was not included in that release.[3]
GIS workflow
[edit]Mustatil's geospatial workflow is intended to preserve geographic information while computer-vision models operate on smaller image tiles. The project documentation describes the following general sequence:[1]
- load a georeferenced raster or other imagery;
- define or import target classes and annotations;
- train or load an object-detection or segmentation model;
- process the source imagery, including tiled processing for large rasters;
- transform retained detections back into the raster's spatial coordinate system;
- visually review candidate detections;
- export accepted results to standard GIS formats.
GeoPackage output is intended to be usable in software such as QGIS. GeoJSON and tabular outputs are also supported by documented workflows.[2]
Editions and versions
[edit]Mustatil is distributed through several channels whose version numbers do not always match. The Python package, Snap package, archived installers and Microsoft Store edition should therefore be treated as separate distribution tracks.
Mustatil 5.6 / Mustatil Legacy
[edit]Mustatil 5.6 is maintained as the free legacy/stable line. The official project website states that the complete 5.6 release remains permanently free, and it is also offered in the Microsoft Store as Mustatil Legacy.[2]
PyPI lists Mustatil 5.6.0, released on 22 June 2026. The package requires Python 3.10 through 3.12 and provides a PySide6/Qt desktop workspace together with detection, annotation, training, remote-sensing, GIS-review and several model-specific plugins.[4]
The Python package can be installed with:
python -m pip install --upgrade mustatil
mustatil
PyPI notes that ordinary package installation cannot automatically select an NVIDIA-specific CUDA build of PyTorch; users who need a particular CUDA configuration may have to install the appropriate PyTorch build separately before installing Mustatil.[4]
Mustatil 6
[edit]Mustatil 6 is the current-generation Microsoft Store edition listed by the official project website. The GitHub release notes state that it became available in the Microsoft Store on 15 July 2026.[3]
The July 2026 project notes describe Mustatil 6 as a newer, still-developing generation and Mustatil 5.6 as the established free version. They also note that some functions could perform differently between the two editions. This distinction is useful when documenting or reproducing a workflow, because a feature demonstrated in one edition should not automatically be assumed to behave identically in the other.[3]
Snap package
[edit]Mustatil is also distributed through the Snap Store for Linux. Snapcraft lists the package name as mustatil and reported latest/stable 1.0.5, last updated on 25 June 2026. This is the Snap package version and should not be confused with the Mustatil 5.6 or Mustatil 6 application naming used by other distribution channels.[5]
Installation:
sudo snap install mustatil
The Snap Store description identifies Mustatil as a GIS AI vision workspace for annotation, AI training, large-image and geospatial detection, web-map viewing, GPKG/GIS export and graphical AI pipelines.[5]
Platform availability
[edit]| Platform / distribution | Availability | Notes |
|---|---|---|
| Windows — Mustatil 6 | Microsoft Store | Current-generation Store edition. |
| Windows — Mustatil Legacy / 5.6 | Microsoft Store and GitHub release | Free legacy/stable edition. |
| Python | PyPI | Mustatil 5.6.0; Python 3.10–3.12 according to PyPI. |
| Linux | Snap Store, Python package and archived Debian package | Snap package is distributed as a separate package track. |
| macOS | Python package and archived macOS package | A macOS package is included in the archived Mustatil 5.6 release on Zenodo/GitHub. |
Scientific archive and citation
[edit]The project is archived on Zenodo as software. The project's website and GitHub README use 10.5281/zenodo.20481110 as the permanent Zenodo reference for the software collection.[2][3]
Individual archived releases have their own Zenodo record DOIs. Examples include:
| Date | Archived material | DOI / record |
|---|---|---|
| 1 June 2026 | An early archived release containing a Mustatil 3.5 installer | 10.5281/zenodo.20481111 |
| 16 June 2026 | Mustatil 5 / 5.1-era archive | 10.5281/zenodo.20722735 |
| 18 June 2026 | Mustatil 5.3 archive | 10.5281/zenodo.20751303 |
| 22 June 2026 | Mustatil 5.6 archive, including Windows, Linux and macOS packages | 10.5281/zenodo.20800096 |
The Zenodo records are software archives and citation records. Their presence on Zenodo should not be confused with peer review or independent scientific validation of the software.[6]
When a specific archived release is used in reproducible research, citing the DOI of that exact version is preferable because it identifies the software state used. The permanent project DOI can additionally be used when referring to Mustatil as a software project in general.
Research use
[edit]The project was created with archaeological remote sensing as a primary use case. Its project paper describes applications to candidate archaeological structures in Saudi Arabia and to other types of landscape features. The same technical workflow can also be used for non-archaeological targets where suitable imagery, training labels and validation procedures exist.[1]
The software is intended to accelerate visual search and documentation rather than replace archaeological or domain-expert interpretation. A machine-learning detection is a prediction and can contain false positives, false negatives, classification mistakes and spatial errors. Chronology, cultural attribution, function and archaeological significance cannot normally be established from an object-detection box alone.[1]
Development
[edit]Mustatil is developed by Tarek Wasfy. The project repository is public on GitHub. The repository contains a large Python/PySide-based codebase and plugin system for the Mustatil 5.x workflow, while the Mustatil 6 release notes describe a new C++ graphical user interface for the Store generation.[3]
The project explicitly discloses the use of generative AI during development. The project paper states that OpenAI ChatGPT was used for code drafting, debugging assistance, documentation and interface-related problem solving, with the author retaining responsibility for the software architecture, workflow decisions, review and final acceptance of generated material.[1]
Licensing
[edit]The GitHub source repository contains the GNU Lesser General Public License version 3 (LGPL-3.0) license text.[7]
Individual bundled frameworks, AI models, datasets and third-party dependencies can have their own licences and usage conditions. Users should therefore check the licences of the specific models and dependencies used in a project, especially when redistributing software, model weights or derived products.
Limitations and reproducibility
[edit]Model performance depends on the training data, imagery, target class, hardware, model family and detection settings. Large-raster processing also involves practical trade-offs between tile size, overlap, confidence thresholds, duplicate suppression, memory consumption and processing time.
The project paper describes Mustatil as local-first, which can be useful for unpublished imagery or sensitive geospatial data, but local execution also means that hardware compatibility, GPU drivers and some machine-learning dependencies remain the user's responsibility.[1]
For reproducible research, a project should record at least:
- the exact Mustatil edition and version;
- the model architecture and model weights;
- model and class configuration;
- confidence and filtering thresholds;
- tile size and overlap;
- source-image version and coordinate-reference system;
- post-processing and manual-review decisions;
- the exported dataset version.
Installation and project links
[edit]Microsoft Store
[edit]- Mustatil 6: Microsoft Store / installer link
- Mustatil Legacy (5.6): Microsoft Store / installer link
Other official project links
[edit]See also
[edit]References
[edit]- 1 2 3 4 5 6 7 8 9 10 11 Mustatil project paper: A local-first geospatial AI workspace for large-raster research surveys, GitHub, dated 27 July 2026.
- 1 2 3 4 5 Mustatil official website, accessed 24 September 2026.
- 1 2 3 4 5 6 7 8 Mustatil GitHub repository, accessed 24 September 2026.
- 1 2 Mustatil on PyPI, accessed 24 September 2026.
- 1 2 Mustatil on the Snap Store, accessed 24 September 2026.
- ↑ Mustatil software archive, Zenodo record 20800096, published 22 June 2026.
- ↑ Mustatil repository license, accessed 24 September 2026.
- provide significant coverage: discuss the subject in detail, not just brief mentions or routine announcements;
- are reliable: from reputable outlets with editorial oversight;
- are independent: not connected to the subject, such as the subject's own website, or sponsored content.
Please add references that meet all three of these criteria. If none exist, the subject is not yet suitable for Wikipedia.