SynthChartNet

SynthChartNet is a multimodal dataset designed for training the SmolDocling model on chart-based document understanding tasks. It consists of 1,981,157 synthetically generated samples, where each image depicts a chart (e.g., line chart, bar chart, pie chart, stacked bar chart), and the associated ground truth is given in OTSL format.

Charts were rendered at 120 DPI using a diverse set of visualization libraries: Matplotlib, Seaborn, and Pyecharts, enabling visual variability in layout, style, and color schemes.

---

Dataset Statistics

• Total samples: 1,981,157

  • Training set: 1,981,157

• Modalities: Image, Text (OTSL format)

• Chart Types: Line, Bar, Pie, Stacked Bar

• Rendering Engines: Matplotlib, Seaborn, Pyecharts

---

Data Format

Each dataset entry is structured as follows:

{
  "images": [PIL Image],
  "texts": [
    {
      "assistant": "<loc_x0><loc_y0><loc_x1><loc_y1><_Chart_>OTSL_REPRESENTATION</chart>",
      "source": "SynthChartNet",
      "user": "<chart>"
    }
  ]
}

---

Intended Use

• Training multimodal models for chart understanding, specifically:

  • Chart parsing and transcription to structured formats (OTSL)

---

Citation

If you use SynthChartNet, please cite:

@article{nassar2025smoldocling,
  title={SmolDocling: An ultra-compact vision-language model for end-to-end multi-modal document conversion},
  author={Nassar, Ahmed and Marafioti, Andres and Omenetti, Matteo and Lysak, Maksym and Livathinos, Nikolaos and Auer, Christoph and Morin, Lucas and de Lima, Rafael Teixeira and Kim, Yusik and Gurbuz, A Said and others},
  journal={arXiv preprint arXiv:2503.11576},
  year={2025}
}
@inproceedings{lysak2023optimized,
  title={Optimized table tokenization for table structure recognition},
  author={Lysak, Maksym and Nassar, Ahmed and Livathinos, Nikolaos and Auer, Christoph and Staar, Peter},
  booktitle={International Conference on Document Analysis and Recognition},
  pages={37--50},
  year={2023},
  organization={Springer}
}