Mohammed Saidul Islam

Automatic evaluators such as reward models play a central role in the alignment and evaluation of large vision–language models (LVLMs). Despite their growing importance, these evaluators are almost exclusively assessed on English-centric benchmarks, leaving open the question of how well these eva...