Menu
March 2024

New publications and test tools

If you missed our presentations at CAA 2023 in Amsterdam or want to test your own machine learning models on our coins, click here.

The publications
•    Gampe S., Tolle K. (2023, December 22): Creating an additional class layer with machine learning to counter overfitting in an unbalanced ancient coin dataset, https://zenodo.org/records/10424274
•    Peter U., Franke C., Köster J., Tolle K., Gampe S., Stolba V.F. (2024, February 8): CORPUS NUMMORUM – A Digital Research Infrastructure for Ancient Coins, https://doi.org/10.5281/zenodo.10633905
had been positivly reviewed by PCI Archaeology and could therefore be published as part of the Proceedings of CAA Amsterdam 2023.
In this context, a documentation about the editor was also created on Zenodo:
•    Köster J., Franke C. (2024, January 4): Corpus Nummorum Editor, https://doi.org/10.5281/zenodo.10458195

All links can also be found on our website under the Resources section.

We also published our Coin Image Dataset (https://zenodo.org/records/10033993) in October. This dataset contains 115,160 coin images from over 29,000 coins. This dataset was used to train the machine learning-based image recognition models in the CN project. With the publication of the dataset, we invite everyone to test and implement their own ideas and models with this large dataset.
The image recognition results we have achieved can also be tested with the coins in CN or with your own images. The Jupyter notebook (Google Colab) for coin type and mint recognition has been uploaded to GitHub: https://github.com/Frankfurt-BigDataLab/IR-on-coin-datasets. The links are also available on our website: https://www.corpus-nummorum.eu/resources/open-source-tools.

The English NLP pipeline can be accessed directly from the coin descriptions. This can also be used to test your own descriptions. Both the subjects and objects (NER) and the relationships that connect them (RE) are labeled: https://github.com/Frankfurt-BigDataLab/NLP-on-multilingual-coin-datasets.https://doi.org/10.5281/zenodo.10458195

Author: Ulrike Peter