rimon-elias/HieroSet

HieroSet: A Hierarchical Image Dataset for Hieroglyphic Gardiner’s List and Ancient Egyptian Deities Description

HieroSet is a dataset developed for research on the automatic detection, recognition, and analysis of Egyptian hieroglyphs and ancient Egyptian deities using computer vision and deep learning techniques.

The dataset contains multiple representations and resources designed to support hierarchical approaches to hieroglyph recognition. It includes datasets, trained models, configuration files, class-name mappings, and auxiliary metadata. The dataset is organized into coarse- and fine-level resources. The coarse-level resources support the detection and recognition of broader hieroglyph categories, while the fine-level resources provide more detailed hieroglyph classes. The dataset also includes a flat representation for experiments that treat the classes without hierarchical grouping.

The accompanying resources include:

• coarse_dataset.zip — coarse-level dataset
• flat_dataset.zip — flat/non-hierarchical dataset
• fine_dataset-part1.zip — first part of the fine-level dataset
• fine_dataset-part2.zip — second part of the fine-level dataset
• fine_dataset-part3.zip — third part of the fine-level dataset
• coarse.pt — trained coarse-level model
• flat.pt — trained flat-level model
• fine_models.zip — fine-level trained models
• coarse_dataset.yaml — coarse-level dataset configuration
• fine_yamls.zip — fine-level dataset configurations
• coarse_names.json — coarse-level class names
• groups.json — hierarchical class-group information
• mapping.json — class/mapping information HieroSet is intended to facilitate reproducible research in Egyptian hieroglyph and deity recognition, object detection, hierarchical classification, and related computer vision applications. Researchers may use the dataset and accompanying resources for experimentation, benchmarking, and development of new computational methods for hieroglyph analysis.

Keywords

Egyptian hieroglyphs; Egyptian deities; HieroSet; hieroglyph recognition; hieroglyph detection; ancient Egyptian; computer vision; deep learning; object detection; image classification; hierarchical classification; digital humanities; cultural heritage; Egyptology Resource Type

Language

English

Copyright and Licensing

HieroSet is provided primarily for research purposes.

The dataset contains images and other materials collected from publicly accessible online sources. Some images may originate from third-party websites, including Pinterest and Roboflow. The copyright status, ownership, and permitted uses of individual images may vary.

The dataset maintainer does not claim ownership of third-party images and does not grant a blanket license to third-party images contained in the dataset. The absence of a license declaration for this repository should not be interpreted as a statement that all images are in the public domain or otherwise free of copyright restrictions.

Where the rights status of particular material is known, users should follow the terms and attribution requirements associated with that material.

If you are a copyright or other rights holder and believe that material in this dataset has been included without appropriate authorization, please contact the dataset maintainer so that the material can be reviewed and, where appropriate, removed or replaced.

Intended Use

HieroSet is intended primarily for research and educational purposes, including:

• Egyptian hieroglyph detection and recognition
• Recognition of ancient Egyptian deities
• Object detection
• Image classification
• Hierarchical classification
• Computer vision research
• Deep learning research
• Benchmarking and experimentation
• Digital humanities and cultural heritage research
• Development of computational methods for Egyptian hieroglyph analysis

Dataset Organization

The dataset provides coarse, fine, and flat representations to support different experimental settings.

The coarse-level resources group hieroglyphs into broader categories and are intended for experiments involving higher-level detection and recognition.

The fine-level resources provide more detailed classes for experiments requiring finer-grained hieroglyph recognition.

The flat representation treats the classes without hierarchical grouping and can be used for experiments that do not require an explicit hierarchical structure.

Models and Auxiliary Resources

The repository also provides trained models, dataset configuration files, class-name mappings, hierarchical group information, and other auxiliary resources intended to facilitate reproducible experimentation.

These resources may be used alongside the corresponding dataset representations for research and benchmarking.

Limitations

The dataset was assembled from heterogeneous sources, and the provenance and licensing status of individual images may not be uniform.

The dataset may also contain variations in image quality, resolution, visual style, background, and representation.

Citation

If you use HieroSet in academic or other research work, please cite the associated publication:

@inproceedings{Elias26, author = {Rimon Elias}, title = {HieroSet: A Hierarchical Image Dataset for Hieroglyphic Gardiner’s List and Ancient Egyptian Deities}, booktitle = {ACM 10th International Conference on Advances in Artificial Intelligence, ICAAI 2026}, publisher = {ACM}, address = {London, UK}, year = {2026} }

and

@misc{rimon_elias_2026, author = { Rimon Elias }, title = { HieroSet }, year = 2026, url = { https://huggingface.co/datasets/rimon-elias/HieroSet }, doi = { 10.57967/hf/10350 }, publisher = { Hugging Face } }