October 2, 2025Open Access

Perception-R1: Advancing Multimodal Reasoning Capabilities of MLLMs via Visual Perception Reward

View Full Paper

Ask AI

Bookmark

Perception-R1 improves multimodal perception and reasoning in MLLMs through a novel visual perception reward.
Extensive experiments reveal that only 1,442 training data are sufficient to achieve state-of-the-art benchmark performance.
Existing RLVR methods inadequately address multimodal perception challenges, limiting reasoning capabilities in MLLMs.
Visual perception rewards were assigned after assessing consistency between generated responses and visual annotations.

Discussion

No takes yet. Share an insight, caveat, or question.

Proposed Perception-R1 enhances multimodal perception and reasoning in MLLMs, suggesting improved training outcomes.

Xiao et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark