See Less, See Right: Bi-directional Perceptual Shaping For Multimodal Reasoning

TL;DR

This paper introduces Bi-directional Perceptual Shaping (BiPS), a training method that enhances multimodal reasoning in vision-language models by improving visual evidence utilization and generalization across domains.

Contribution

BiPS is a novel training approach that uses bidirectional visual cues and consistency constraints to improve visual evidence reliance and domain generalization in VLMs.

Findings

BiPS improves Qwen2.5-VL-7B performance by 8.2% on average across benchmarks.

BiPS enhances out-of-domain generalization to unseen datasets and image types.

BiPS effectively encourages models to rely on fine-grained visual evidence.

Abstract

Large vision-language models (VLMs) often benefit from intermediate visual cues, either injected via external tools or generated as latent visual tokens during reasoning, but these mechanisms still overlook fine-grained visual evidence (e.g., polylines in charts), generalize poorly across domains, and incur high inference-time cost. In this paper, we propose Bi-directional Perceptual Shaping (BiPS), which transforms question-conditioned masked views into bidirectional where-to-look signals that shape perception during training. BiPS first applies a KL-consistency constraint between the original image and an evidence-preserving view that keeps only question-relevant regions, encouraging coarse but complete coverage of supporting pixels. It then applies a KL-separation constraint between the original and an evidence-ablated view where critical pixels are masked so the image no longer supports the original answer, discouraging text-only shortcuts (i.e., answering from text alone) and enforcing fine-grained visual reliance. Across eight benchmarks, BiPS boosts Qwen2.5-VL-7B by 8.2% on average and shows strong out-of-domain generalization to unseen datasets and image types.