Back to Research papers
Research paper index

RefineSVG: Visual Feedback-Driven Reinforcement Learning for Image-to-SVG Generation

Shaobo Liu, Feiqiao Mao, Shuaishuai Zhou, Yan Zhan, Weiqi Tan, Zhiqiong Lu, Zhengping Liang

arXiv:2607.27699Published July 30, 20260 citations
  • cs.CV
  • cs.AI
  • action
  • reinforcement learning

Abstract

We propose RefineSVG, a single-step closed-loop visual feedback framework that enables multimodal large language models (MLLMs) to perform high-fidelity image-to-SVG generation through self-correction. Existing MLLM-based approaches rely on single-pass open-loop inference, where the model receives visual input only once and must generate thousands of SVG code tokens without intermediate verification. This paradigm inevitably leads to geometric drift, error accumulation, and visual hallucination on complex images. RefineSVG overcomes this limitation by invoking an external rendering engine after an initial SVG generation pass to compare the rendered output against the target image. The comparison yields a multi-dimensional visual residual map (Diff-Map) that is fed back to the model as a ReAct-style correction signal, driving a targeted correction step. To support this render-observe-correct interaction, we further introduce an SVG-oriented semantic vocabulary that compresses token sequences by over 52%. A progressive training pipeline spanning supervised fine-tuning, rejection-sampling cold-start data construction, and end-to-end agentic reinforcement learning aligns the model with closed-loop visual correction. Extensive experiments show that RefineSVG consistently outperforms existing baselines in reconstruction fidelity, structural accuracy, and code efficiency.Code is available at https://github.com/liuxiaobo66/RefineSVG.

Read the original paper

This page indexes public paper metadata. The manuscript remains with its original publisher and authors.