Visual-ERM: Reward Modeling for Visual Equivalence

TL;DR

This paper introduces Visual-ERM, a multimodal reward model that provides fine-grained visual feedback for vision-to-code tasks, improving reinforcement learning performance and establishing a new benchmark for visual equivalence evaluation.

Contribution

The paper presents Visual-ERM, a novel reward model that enhances vision-to-code RL by capturing detailed visual discrepancies and introduces VC-RewardBench, a benchmark for structured visual data comparison.

Findings

Visual-ERM improves RL performance on chart-to-code (+8.4) and other visual parsing tasks.

Visual-ERM outperforms existing reward methods in fine-grained visual discrepancy detection.

The benchmark VC-RewardBench effectively evaluates visual equivalence in structured visual data.

Abstract

Vision-to-code tasks require models to reconstruct structured visual inputs, such as charts, tables, and SVGs, into executable or structured representations with high visual fidelity. While recent Large Vision Language Models (LVLMs) achieve strong results via supervised fine-tuning, reinforcement learning remains challenging due to misaligned reward signals. Existing rewards either rely on textual rules or coarse visual embedding similarity, both of which fail to capture fine-grained visual discrepancies and are vulnerable to reward hacking. We propose Visual Equivalence Reward Model (Visual-ERM), a multimodal generative reward model that provides fine-grained, interpretable, and task-agnostic feedback to evaluate vision-to-code quality directly in the rendered visual space. Integrated into RL, Visual-ERM improves Qwen3-VL-8B-Instruct by +8.4 on chart-to-code and yields consistent gains on table and SVG parsing (+2.7, +4.1 on average), and further strengthens test-time scaling via reflection and revision. We also introduce VisualCritic-RewardBench (VC-RewardBench), a benchmark for judging fine-grained image-to-image discrepancies on structured visual data, where Visual-ERM at 8B decisively outperforms Qwen3-VL-235B-Instruct and approaches leading closed-source models. Our results suggest that fine-grained visual reward supervision is both necessary and sufficient for vision-to-code RL, regardless of task specificity.