AI RESEARCH

LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models

arXiv CS.CV

ArXi:2505.15616v2 Announce Type: replace Multimodal Large Language Models (MLLMs) have achieved significant advances in integrating visual and linguistic information, yet their ability to reason about complex and real-world scenarios remains limited. The existing benchmarks are usually constructed in the task-oriented manner without guarantee that different task samples come from the same data distribution, thus they often fall short in evaluating the synergistic effects of lower-level perceptual capabilities on higher-order reasoning.