Research map / Vision-language and multimodal / Multimodal LLMs
Effective Training Data Synthesis for Improving MLLM Chart Understanding
Yuwei Yang, Zeyu Zhang, Yunzhong Hou, Zhuowan Li, Gaowen Liu, Ali Payani, Yuan-Sen Ting, Liang Zheng
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 ECD | LLaVA-Next-Llama3-8B · CharXiv Avg | accuracy: 35.06 -> 51.60 | Improved |
| fine-tune on ECD | LLaVA-Next-Llama3-8B · ChartX | accuracy: 27.69 -> 46.61 | Improved |
| fine-tune on ECD | Qwen2.5-VL-7B · CharXiv Reasoning | accuracy: 41.20 -> 40.20 | Hurt |
| fine-tune on ECD | MiniCPM-V2.6 · ChartBench Binary | accuracy: 75.67 -> 70.77 | Hurt |
| fine-tune on ECD | Qwen2.5-VL-7B · ECDBench Avg | accuracy: 38.19 -> 50.86 | Improved |
| fine-tune on ChartBench instead of baseline | LLaVA-Next-Llama3-8B · CharXiv Avg | accuracy: 35.06 -> 32.86 | Hurt |
| fine-tune on ReachQA | LLaVA-Next-Llama3-8B · CharXiv Avg | accuracy: 35.06 -> 30.68 | Hurt |
| add visual diversification to data synthesis (ablation) | ECD pipeline · FID vs CharXiv | FID: 80.38 -> 60.74 | Improved |
| add image quality filtering (ablation) | ECD pipeline · FID vs CharXiv | FID: 66.20 -> 60.74 | Improved |
| add QA filtering (ablation) | ECD fine-tuning · CharXiv Avg | accuracy: 46.64 -> 47.02 | Improved |
| increase chart types from 5 to 29 (ablation) | ECD · CharXiv Avg | accuracy | Improved |
| use reasoning questions only vs descriptive only (ablation) | ECD · CharXiv Avg | accuracy: +4.16% vs +6.40% | Mixed |
Closest papers by meaning
- EvoChart: A Benchmark and a Self-Training Approach Towards Real-World Chart Understanding (AAAI 2025 · similarity 0.91)
- CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs (NeurIPS 2024 · similarity 0.90)
- ChartNet: A Million-Scale, High-Quality Multimodal Dataset for Robust Chart Understanding (CVPR 2026 · similarity 0.89)
- Synthesize Step-by-Step: Tools Templates and LLMs as Data Generators for Reasoning-Based Chart VQA (CVPR 2024 · similarity 0.88)
- Chart-FR1: Visual Focus-Driven Fine-Grained Reasoning on Dense Charts (CVPR 2026 · similarity 0.87)
- ChartSketcher: Reasoning with Multimodal Feedback and Reflection for Chart Understanding (NeurIPS 2025 · 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.
