Research map / Vision-language and multimodal / Multimodal LLMs
Render-in-the-Loop: Vector Graphics Generation via Visual Self-Feedback
Guotao Liang, Zhangcheng Wang, Juncheng Hu, Haitao Zhou, Ziteng Xue, Jing Zhang, Dong Xu, Qian Yu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Visual Self-Feedback training (ablation) | Qwen3-VL-8B SFT · MMSVG-Icon | FID: 148.72 -> 127.98 | Improved |
| add Visual Self-Feedback training (ablation) | Qwen3-VL-8B SFT · MMSVG-Illustration | FID: 155.84 -> 137.86 | Improved |
| add Visual Self-Feedback training (ablation) | Qwen3-VL-8B SFT · MMSVG-Icon | LPIPS: 0.312 -> 0.188 | Improved |
| add Render-and-Verify at inference (ablation) | Ours (VSF) · MMSVG-Icon | FID: 127.98 -> 127.64 | Improved |
| add Render-and-Verify at inference (ablation) | Ours (VSF) · MMSVG-Icon | LPIPS: 0.188 -> 0.172 | Improved |
| add Render-and-Verify at inference (ablation) | Ours (VSF) · MMSVG-Illustration | LPIPS: 0.193 -> 0.178 | Improved |
| add Render-and-Verify at inference (ablation) | Ours (VSF) · MMSVG-Illustration | FID: 137.86 -> 137.79 | No clear effect |
| add naive multi-turn visual feedback without training (ablation) | GPT-5 · MMSVG-Illustration | LPIPS: 0.345 -> 0.388 | Hurt |
| add naive multi-turn visual feedback without training (ablation) | GPT-5 · MMSVG-Icon | LPIPS: 0.387 -> 0.405 | Hurt |
| add naive multi-turn visual feedback without training (ablation) | Qwen3-VL / GPT-5 · MMSVG-Icon | CLIP: 0.311 -> 0.294 | Hurt |
| replace OmniSVG with Render-in-the-Loop (0.85M vs 2M data) | Ours (VSF+RaV) · MMSVG-Icon | FID: 130.56 -> 127.64 | Improved |
| replace InternSVG (16M data) with Render-in-the-Loop | Ours (VSF+RaV) · MMSVG-Icon | FID: 128.80 -> 127.64 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- InternSVG (8B) — InternSVG: Towards Unified SVG Tasks with Multimodal Large Language Models (ICLR 2026) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rendering-Aware Reinforcement Learning for Vector Graphics Generation (NeurIPS 2025 · similarity 0.91)
- Chat2SVG: Vector Graphics Generation with Large Language Models and Image Diffusion Models (CVPR 2025 · similarity 0.88)
- StarVector: Generating Scalable Vector Graphics Code from Images and Text (AAAI 2025 · similarity 0.88)
- VIGC: Visual Instruction Generation and Correction (AAAI 2024 · similarity 0.87)
- Empowering LLMs to Understand and Generate Complex Vector Graphics (CVPR 2025 · similarity 0.87)
- VisCodex: Unified Multimodal Code Generation via Merging Vision and Coding Models (ICLR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
