94 Quantifying and Learning Linear Symmetry-Based Disentanglement (LSBD) 4.8.1 The Challenge The challenge evolved around a dataset consisting of digitised versions of 938 archaeological objects from the Josefina Ramos de Cox Museum in Peru with varied geometry and artistic styles. There were two different retrieval challenges focusing on either shape or culture. For the retrieval-by-shape challenge, archaeologists in the museum classified the scanned objects by shape using specific taxonomies for archaeological artefacts, resulting in eight categories (see Figure 4.14). Given a query object, the goal is to retrieve objects from the same category. This is considered the easier challenge, since the shape category links directly to geometric properties of the objects, for which suitable methods already exist in the 3D shape retrieval literature. Nevertheless, in some cases the distinction between objects in different classes is barely perceivable. Figure 4.14: Sample objects for every class of the retrieval-by-shape dataset. Image from Sipiran et al. (2021). In the retrieval-by-culture challenge, 637 out of the total 938 objects are classified into six categories based on the culture these objects come from (see Figure 4.15. This makes for a more difficult challenge, because the categorisation of objects is based on semantic meaning rather than direct geometric properties. In particular, models from the same culture can have varied shapes, and the most distinguishable characteristic is probably the combination of geometry and painting style.
RkJQdWJsaXNoZXIy MjY0ODMw