Research map / Large language models / Agents, code and math
VisCoder2: Building Multi-Language Visualization Coding Agents
Yuansheng Ni, Songcheng Cai, Xiangchao Chen, Jiarong Liang, Zhiheng Lyu, Jiaqi Deng, Kai Zou, Ping Nie, Fei Yuan, Xiang Yue, Wenhu Chen
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 |
|---|---|---|---|
| fine-tune on VisCode-Multi-679K | Qwen2.5-Coder-7B · VisPlotBench overall | Execution pass rate (%): 51.2 -> 70.9 | Improved |
| fine-tune on VisCode-Multi-679K | Qwen2.5-Coder-32B · VisPlotBench overall | Execution pass rate (%): 57.5 -> 73.1 | Improved |
| add multi-round self-debug (up to 3 rounds) | VisCoder2-32B · VisPlotBench overall | Execution pass rate (%): 73.1 -> 82.4 | Improved |
| add self-debug | GPT-4.1 · VisPlotBench LaTeX | Execution pass rate (%): 31.3 -> 66.1 | Improved |
| add CoSyn synthetic data (ablation) | Qwen2.5-Coder-7B · VisPlotBench LilyPond | Execution pass rate (%): 5.5 -> 65.5 | Improved |
| train only on The-Stack-V2 subset (ablation) | Qwen2.5-Coder-7B · VisPlotBench overall | Execution pass rate (%): 51.2 -> 49.0 | Hurt |
| train only on StarVector SVG subset (ablation) | Qwen2.5-Coder-7B · VisPlotBench overall | Execution pass rate (%): 51.2 -> 40.1 | Hurt |
| add Code-Feedback multi-turn dialogues (ablation) | Qwen2.5-Coder-7B · VisPlotBench overall, self-debug | Execution pass rate (%): 59.0 -> 63.1 | Improved |
| combine all four data sources (ablation) | Qwen2.5-Coder-7B · VisPlotBench overall | Execution pass rate (%): 59.2 -> 70.9 (best single source CoSyn) | Improved |
| fine-tune on VisCode-Multi-679K | Qwen2.5-Coder-32B · VisPlotBench SVG | Execution pass rate (%): 93.9 -> 81.5 | Hurt |
Closest papers by meaning
- VisCodex: Unified Multimodal Code Generation via Merging Vision and Coding Models (ICLR 2026 · similarity 0.91)
- JanusCoder: Towards a Foundational Visual-Programmatic Interface for Code Intelligence (ICLR 2026 · similarity 0.90)
- VisInteract: Towards Dynamic Interactive Text-to-Visualization under Imperfect Queries (NeurIPS 2026 · similarity 0.88)
- VIGC: Visual Instruction Generation and Correction (AAAI 2024 · similarity 0.87)
- VCG-Bench: Towards A Unified Visual-Centric Benchmark for Structured Generation and Editing (ICML 2026 · similarity 0.87)
- Generative Visual Code Mobile World Models (ICML 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.
