Research map / Vision-language and multimodal / Multimodal LLMs
Reliable Reasoning in SVG-LLMs via Multi-Task Multi-Reward Reinforcement Learning
Haomin Wang, Qi Wei, Qianli Ma, Shengyuan Ding, Jinhui Yin, Kai Chen, hongjie Zhang
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 chain-of-thought reasoning to SFT+RL model (ablation) | CTRL-S (SFT+RL) · SArena-Icon Text-to-SVG | FID: 12.161 -> 11.584 | Improved |
| add chain-of-thought reasoning to SFT model (ablation) | CTRL-S (SFT) · SArena-Icon Text-to-SVG | success rate (%): 85.75 -> 92.02 | Improved |
| add chain-of-thought reasoning to SFT model (ablation) | CTRL-S (SFT) · SArena-Icon Image-to-SVG | success rate (%): 81.96 -> 90.12 | Improved |
| add chain-of-thought to SFT+RL model (token cost) (ablation) | CTRL-S (SFT+RL) · SArena-Icon Text-to-SVG | output tokens: 274 -> 346 | Hurt |
| add RL post-training after SFT | CTRL-S · SArena-Icon Text-to-SVG | FID: 18.154 -> 11.584 | Improved |
| add RL post-training after SFT | CTRL-S · SArena-Icon Text-to-SVG | output tokens: 1.1k -> 346 | Improved |
| add RL post-training after SFT | CTRL-S · SVG-Sophia Code Refinement | success rate (%): 84.37 -> 99.79 | Improved |
| add Long-CLIP image-text similarity reward (ablation) | CTRL-S RL · SArena-Icon Text-to-SVG | CLIP-T2I: 24.573 -> 25.444 | Improved |
| add code efficiency reward (ablation) | CTRL-S RL · SArena-Icon Text-to-SVG | output tokens: 701 -> 346 | Improved |
| add code efficiency reward (ablation) | CTRL-S RL · SArena-Icon Text-to-SVG | inference time (s/svg): 7.121 -> 4.439 | Improved |
| use reward ratio 2:1:1 instead of 1:1:1 (ablation) | CTRL-S RL · SArena-Icon Text-to-SVG | FID: 12.575 -> 11.584 | Improved |
| use reward ratio 2:1:1 instead of 5:3:1 (ablation) | CTRL-S RL · SArena-Icon Text-to-SVG | CLIP-T2I: 24.653 -> 25.944 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- InternSVG-8B — InternSVG: Towards Unified SVG Tasks with Multimodal Large Language Models (ICLR 2026) · better on 4 of 5 shared results
- LIVE — LIVE: Learnable In-Context Vector for Visual Question Answering (NeurIPS 2024) · better on 1 of 3 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 0.90)
- InternSVG: Towards Unified SVG Tasks with Multimodal Large Language Models (ICLR 2026 · similarity 0.88)
- Empowering LLMs to Understand and Generate Complex Vector Graphics (CVPR 2025 · similarity 0.87)
- Breaking the SFT Plateau: Multimodal Structured Reinforcement Learning for Chart-to-Code Generation (ICLR 2026 · similarity 0.87)
- Perception-R1: Advancing Multimodal Reasoning Capabilities of MLLMs via Visual Perception Reward (ICLR 2026 · similarity 0.86)
- Visual Jigsaw Post-Training Improves MLLMs (ICLR 2026 · similarity 0.86)
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.
