OmniDiagram: Advancing Unified Diagram Code Generation via Visual Interrogation Reward

2026-04-07

AI Summary

Existing approaches to chart code generation are constrained by task-specific formulations and programming languages, limiting their ability to support diverse chart types and structured visualization requirements. This work proposes OmniDiagram, a unified framework that integrates multilingual chart code and task definitions, and introduces the novel Visual Interrogation Verifies All (Viva) mechanism. Viva replaces conventional syntactic or pixel-level matching with generative visual interrogation to provide fine-grained visual fidelity feedback, enabling self-evolving training without human annotations. Leveraging Viva, the framework combines supervised fine-tuning and reinforcement learning to optimize structural alignment of rendered outputs. Experiments demonstrate that OmniDiagram achieves new state-of-the-art performance across multiple benchmarks and introduces M3²Diagram, the first large-scale chart code dataset, substantially improving generation quality in cross-chart-type and cross-language scenarios.

Abstract

The paradigm of programmable diagram generation is evolving rapidly, playing a crucial role in structured visualization. However, most existing studies are confined to a narrow range of task formulations and language support, constraining their applicability to diverse diagram types. In this work, we propose OmniDiagram, a unified framework that incorporates diverse diagram code languages and task definitions. To address the challenge of aligning code logic with visual fidelity in Reinforcement Learning (RL), we introduce a novel visual feedback strategy named Visual Interrogation Verifies All (\textsc{Viva}). Unlike brittle syntax-based rules or pixel-level matching, \textsc{Viva} rewards the visual structure of rendered diagrams through a generative approach. Specifically, \textsc{Viva} actively generates targeted visual inquiries to scrutinize diagram visual fidelity and provides fine-grained feedback for optimization. This mechanism facilitates a self-evolving training process, effectively obviating the need for manually annotated ground truth code. Furthermore, we construct M3$^2$Diagram, the first large-scale diagram code generation dataset, containing over 196k high-quality instances. Experimental results confirm that the combination of SFT and our \textsc{Viva}-based RL allows OmniDiagram to establish a new state-of-the-art (SOTA) across diagram code generation benchmarks.

Problem

• diagram code generation
• visual fidelity
• unified framework
• reinforcement learning
• programmable diagram

Innovation

• Unified Diagram Code Generation
• Visual Interrogation Reward
• Reinforcement Learning
• Self-evolving Training
• Large-scale Dataset