Riedones3D: a celtic coin dataset for registration and fine-grained clustering

Abstract

Clustering coins with respect to their die is an important component of numismatic research and crucial for understanding the economic history of tribes (especially when literary production does not exist, in celtic culture). It is a very hard task that requires a lot of times and expertise. To cluster thousands of coins, automatic methods are becoming necessary. Nevertheless, public datasets for coin die clustering evaluation are too rare, though they are very important for the development of new methods. Therefore, we propose a new 3D dataset of 2 070 scans of coins. With this dataset, we propose two benchmarks, one for point cloud registration, essential for coin die recognition, and a benchmark of coin die clustering. We show how we automatically cluster coins to help experts, and perform a preliminary evaluation for these two tasks. The code of the baseline and the dataset will be publicly available at https://www.npm3d.fr/coins-riedones3d and https: //www.chronocarto.eu/spip.php?article84&lang=fr.

Unsupervised Statistical Learning for Die Analysis in Ancient Numismatics

“…Previous to the above mentioned pre-sorting software CADS [7] and the supervised clustering in [8], computer vision techniques in numismatics have been used for tasks such as ancient coin classification by numismatic type or issue [16]- [21], coin identification [22], and high-volume type classification of contemporary coins [23]. While type classification is an interesting computer vision problem, it only is of limited value for historical research, because most available coin images already exist in the form of labelled data (Sec.…”

“…The Riedones3D dataset of Horache et al [8] comprises of 3D-scans of a hoard find of 1 035 ancient Celtic coins. The coins were struck from a small number of dies, which is common for hoards of freshly minted coins that did not circulate much.…”

“…4.3 for details on differences between our approach and CADS). The difficulty of computer assisted die studies is further underlined by Horache et al [8], who consider supervised deep representation clustering (based on [9]) for 3D-scans of a hoard find of Celtic coinage. The challenges of creating such a high-quality dataset lie not so much in the availability of a 3D-scanning device, as in the access to the coinage for scanning.…”

“…1(a) versus 1(b), cannot be recovered (see Sec. 6.2 for further details on quantitative and conceptual differences of [8] to our approach). To the best of our knowledge, besides our proposal there is currently no unsupervised method capable of end-to-end clustering of coins by dies.…”

“…In the Bayesian framework, the number of dies used to strike the sample, a key unknown [15], is an object of inference, to be predicted in conjunction with the die labels. Its probabilistic foundation sets it apart from other distance-based clustering algorithms, such as hierarchical clustering employed by CADS, the graph-based clustering of [8], or partitional approaches like k-means. The use of a microclustering prior addresses the fact that in ancient numismatics the frequency of surviving coins per die is generally small, which results in relatively small class sizes even for large samples.…”