ChartGalaxy: A Dataset for Infographic Chart Understanding and Generation

Zhen Li, Duan Li, Yukai Guo, Xinyuan Guo, Bowen Li, Lanxi Xiao, Shenyu Qiao, Jiashu Chen, Zijian Wu, Hui Zhang, Xinhuan Shu, Shixia Liu

Open source

Type: paper
Venue: ICLR 2026
Year: 2026
Source: huggingface
Access: public
Language: en
Added: 2026-09-29
Verified: 2026-09-29

Summary

Introduces ChartGalaxy, a dataset of 1,880,687 infographic charts (1.82M synthetic + 61,978 real) across 75 chart types, 440 variations, and 68 layout templates, each paired with its tabular source data, for infographic chart understanding and generation research.

Research notes

Key findings: Scale: 1,880,687 infographic charts mixing programmatic synthetic charts (1,818,709) with real charts from 18 websites (61,978); Covers 75 chart types, 440 variations, and 68 layout templates; Every chart is paired with its tabular source data, enabling chart-to-table, code generation, and layout-aware editing tasks
OpenReview forum id P4lFbvZ4HH. Associated dataset: ChartGalaxy/ChartGalaxy (same batch).