CAVE: A Structured Credit Assignment Approach for Fragmented Visual Evidence Reasoning

2026-05-13

AI Summary

This work addresses the limited reasoning capability of vision-language models when handling non-local and semantically ambiguous fragmented visual evidence. To this end, we propose CAVE, the first approach to introduce structured credit assignment into fragmented visual reasoning. Built upon the GRPO reinforcement learning framework, CAVE leverages three types of process signals—belief updating, evidence acquisition, and adaptive focus control—to evaluate the contribution of intermediate reasoning steps at the action level, thereby enabling process-level supervision and intermediate evidence guidance. We further construct the TRACER-Bench benchmark to provide fine-grained supervisory signals for such reasoning. Experiments demonstrate that CAVE significantly improves performance on both public benchmarks and TRACER-Bench in fragmented visual reasoning, while maintaining competitive results on general multimodal tasks, showcasing enhanced robustness and deeper reasoning capabilities.

Abstract

Vision-Language Models (VLMs) have achieved strong performance on general multimodal reasoning, yet remain challenged in integrating nonlocal visual information to support semantically underdetermined visual reasoning. We describe this challenge as Fragmented Visual Reasoning. To this end, we propose Credit Assignment for Visual Evidence (CAVE), a structured process-reward method based on GRPO for interleaved visual reasoning. Specifically, CAVE evaluates the contribution of intermediate steps at the action level via three complementary reasoning process signals: belief update, evidence acquisition, and adaptive focus control, thereby guiding the model to optimize each reasoning action and learn more reliable visual reasoning strategies. Meanwhile, we construct TRACER-Bench, which covers four nonlocal and semantically confusable reasoning dimensions and provides key intermediate evidence to supervise reasoning paths. Experiments demonstrate that CAVE substantially improves performance on tasks requiring fragmented visual evidence integration, covering both public benchmarks and our newly introduced TRACER-Bench, while retaining competitive performance on general multimodal evaluations. Further analyses reveal that CAVE effectively improves the visual reasoning capacity and exhibits stronger robustness under longer-range and deeper cross-region dependencies.

Problem

• Fragmented Visual Reasoning
• Vision-Language Models
• Nonlocal Visual Information
• Semantically Underdetermined Reasoning
• Visual Evidence Integration

Innovation

• Credit Assignment
• Fragmented Visual Reasoning
• Process Reward
• Visual-Language Models
• Structured Reasoning