Be Consistent! Enhancing Robust Visual Reasoning in LVLMs with Consistency Constraints
For researchers building reliable vision-language systems, this work addresses the underexplored problem of logical consistency in complex visual reasoning, providing both a benchmark and a training method.
LVLMs struggle with logical consistency in complex visual reasoning. The authors introduce ConVBench, a benchmark with logically equivalent question pairs, and ConVLM, a GRPO-based method with a consistency reward that improves both accuracy and consistency, achieving robust gains without strict supervision.
While Large Vision-Language Models (LVLMs) exhibit strong perceptual capabilities, they remain vulnerable in visual reasoning tasks. Existing benchmarks largely focus on symbolic mathematical or scientific problems and simple vision-centric tasks, offering limited assessment of complex visual reasoning and logical consistency, a critical requirement for reliable reasoning systems. We introduce ConVBench, a complex vision-centric reasoning benchmark in which each image is paired with two logically equivalent questions across six categories: action and state, complex counting, spatial reasoning, causal and intent understanding, commonsense reasoning, and temporal perception. To complement this benchmark, we define two evaluation metrics, logical consistency and robust accuracy, that jointly assess both the correctness and consistency of model responses. We further present ConVLM, which improves LVLM reasoning through Group Relative Policy Optimization (GRPO)-based reinforcement learning with a novel consistency reward. This method leverages automatically generated logically equivalent question-answer pairs and a dual-reward design combining accuracy- and consistency-based signals, encouraging agreement between paired responses. The framework functions effectively with or without strict answer supervision.