Validate retrieval on large-scale, multi-institutional collections

Establish the efficacy of the automated goldsmith’s-mark retrieval pipeline on large-scale, multi-institutional datasets containing photographs of gold- and silverwork and their marks.

Background

The evaluation uses only 146 query images from a relatively small dataset, so the authors characterize the work as a pilot study. They note that the system is not yet sufficiently validated for provenance research involving object transitions across multiple collections. Testing the method across larger collections and institutions is therefore an explicitly unresolved empirical question relevant to practical deployment.