Evaluating Object-Centric Models beyond Object Discovery

Krishnakant Singh, Simone Schaub-Meyer, Stefan Roth · 2026

Read Evaluating Object-Centric Models beyond Object Discovery by Krishnakant Singh, Simone Schaub-Meyer, Stefan Roth (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07532.txt.

Description

Object-centric models promise scene representations that generalize compositionally, yet benchmarks usually test only object discovery and simple probing. Krishnakant Singh and two co-authors argue those tests say little about whether the representations are useful.

They benchmark object-centric models with instruction-tuned vision-language models as evaluators across diverse visual question answering data sets, measuring how well the representations support complex thinking. They also present a unified task and metric that judges localization and representation usefulness instead of separately. Representation learning teams get an testing that finally matches the goals object-centric learning claims.

Catalog record

CREATORKrishnakant Singh, Simone Schaub-Meyer, Stefan Roth
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07532

Browse the full library · About the Tower of Babel