What can Computer Vision learn from Ranganathan?

Mayukh Bagchi, a researcher at the University of Trento, Italy, recently presented a novel approach (joint work with Professor Fausto Giunchiglia, University of Trento, Italy) to solving the Semantic Gap Problem (SGP) in Computer Vision (CV). Speaking at the Documentation Research and Training Centre (DRTC),  Indian Statistical Institute, Bangalore – a key partner in the DataScientia initiative – Bagchi explored how the classification principles of S.R. Ranganathan, suitably adapted, can transform modern CV dataset design and construction. The presentation focused on the critical intersection of Knowledge Organization (KO) and Knowledge Representation (KR), arguing that the flaws in current CV benchmarks like ImageNet stem from an underlying unprincipled annotation process.

By adapting traditional KO principles to digital KR, the proposed vTelos methodology provides a systematic framework for image annotation and CV dataset design. This approach reduces subjective classification errors and ensures a more precise alignment between visual data and linguistic labels. The talk demonstrated that returning to the foundations of information science and faceted classification is essential for building the next generation of robust, high-accuracy CV datasets and systems.

Useful links

• Research Paper: https://arxiv.org/pdf/2601.22634