keyfinder08/EpiSAM_Inscriptions_Project

EpiSAM: Weights

Project Website

For EpiSAM dataset refer to the project website.

We provide two weights:

• episam.pth is the checkpoint for the segmentation module.
• 'binarizer.pth' is the checkpoint for the binarizer used in the self-prompting module. For more details on this refer to the second paper listed in citations section below.

Training code is provided in the GitHub repository, which can be found from the project website.

Citation

If you use our dataset in your research, please cite the following papers:

@InProceedings{10.1007/978-3-032-36039-7_18,
author="Sharma, Arnav
and Jena, Pratyush
and Joseph, Amal
and Sarvadevabhatla, Ravi Kiran",
title="EpiSAM: Character Segmentation in Challenging Stone Inscriptions",
booktitle="Document Analysis and Recognition -- ICDAR 2026",
year="2027",
publisher="Springer Nature Switzerland",
address="Cham",
pages="299--315",
isbn="978-3-032-36039-7"
}


@inproceedings{Jena_2025,
series={ICVGIP 2025},
title={Unveiling Text in Challenging Stone Inscriptions: A Character-Context-Aware Patching Strategy for Binarization},
url={http://dx.doi.org/10.1145/3774521.3774539},
DOI={10.1145/3774521.3774539},
booktitle={Proceedings of the Sixteen Indian Conference on Computer Vision, Graphics and Image Processing},
publisher={ACM},
author={Jena, Pratyush and Joseph, Amal and Sharma, Arnav and Sarvadevabhatla, Ravi Kiran},
year={2025},
month=Dec,
pages={1–9},
collection={ICVGIP 2025}
}

Acknowledgement

We sincerely acknowledge The Mythic Society Bengaluru for providing the inscription images used in this work. These resources are part of the Inscriptions 3D Digital Conservation Project, an initiative aimed at preserving and digitizing valuable epigraphic heritage.

For more information about the project, please visit:

Akshara Bhandara – Inscriptions 3D Digital Conservation Project.

---