Search papers, labs, and topics across Lattice.
The paper introduces Adaptive Visual Reasoning (AVR), a framework that decomposes visual reasoning into perception, logical reasoning, and answer application, enabling VRMs to dynamically select efficient reasoning formats. To train AVR, the authors adapt Group Relative Policy Optimization (FS-GRPO) to encourage the selection of the most efficient reasoning format without sacrificing correctness. Experiments on vision-language benchmarks demonstrate that AVR reduces token usage by 50-90% while maintaining accuracy, particularly in perception-heavy tasks, effectively mitigating overthinking in VRMs.
Visual reasoning models can slash token usage by up to 90% without sacrificing accuracy by adaptively choosing reasoning paths based on task complexity.
Visual reasoning models (VRMs) have recently shown strong cross-modal reasoning capabilities by integrating visual perception with language reasoning. However, they often suffer from overthinking, producing unnecessarily long reasoning chains for any tasks. We attribute this issue to Reasoning Path Redundancy in visual reasoning: many visual questions do not require the full reasoning process. To address this, we propose AVR, an adaptive visual reasoning framework that decomposes visual reasoning into three cognitive functions: visual perception, logical reasoning, and answer application. It further enables models to dynamically choose among three response formats: Full Format, Perception-Only Format, and Direct Answer. AVR is trained with FS-GRPO, an adaptation of Group Relative Policy Optimization that encourages the model to select the most efficient reasoning format while preserving correctness. Experiments on multiple vision-language benchmarks show that AVR reduces token usage by 50--90\% while maintaining overall accuracy, especially in perception-intensive tasks. These results demonstrate that adaptive visual reasoning can effectively mitigate overthinking in VRMs. Code and data are available at: https://github.com/RunRiotComeOn/AVR.
