MatSciFig

MatSciFig is a large-scale dataset of panel-level crops from scientific compound figures in materials science literature, paired with panel-specific subcaptions and extended summaries extracted from the source papers.

Compound figures - multi-panel images labeled A, B, C… ; are the dominant figure type in scientific publications. MatSciFig breaks these figures into individual panels and links each panel to its corresponding textual description, enabling fine-grained vision-language research on scientific imagery.

---

Dataset Statistics

	Count
Source figures	~180,571
Panels with full metadata	391,606

---

Columns

Column	Type	Description
image	bytes	Panel crop in original format
image_id	string	Parent figure identifier
panel_suffix	string	Panel label (A, B, C … or single)
visualization_category	string	High-level category (e.g. Microscopy, Generic Plot, Photograph)
visualization_subtype	string	Fine-grained subtype (e.g. SEM, Contour Plot, XRD Pattern)
subcaption	string	Panel-level caption from the source paper
summary	string	Extended description of the panel content

---

Example

from datasets import load_dataset

ds = load_dataset("subham2507/MatSciFig")
sample = ds["train"][0]

print(sample["panel_suffix"])            # "A"
print(sample["visualization_category"])  # "Microscopy"
print(sample["visualization_subtype"])   # "SEM"
print(sample["subcaption"])              # "SEM image of Fe2O3 nanorods..."
print(sample["summary"])                 # "The SEM image reveals..."
sample["image"].show()                   # displays the panel crop

---

Data Collection

• Compound figures collected from open-access materials science publications
• Panels detected and segmented using a custom-trained YOLO-based compound figure detector
• Subcaptions and summaries extracted and aligned from paper text using automated parsing
• Only panels with complete metadata are included

---

Intended Use

• Scientific figure understanding and captioning
• Vision-language model training and evaluation
• Materials science multimodal research
• Panel classification by visualization type

---

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

• ✅ Free to use for research and non-commercial purposes
• ✅ Share and adapt with attribution
• ❌ Commercial use is not permitted

---

Citation

If you use MatSciFig in your research, please cite:

@article{ghosh2026unlocking,
  title={Unlocking the Visual Record of Materials Science: A Large-Scale Multimodal Dataset from Scientific Literature},
  author={Ghosh, Subham and Tiwari, Shubham and Ibrahim, Mohammad and Tewari, Abhishek},
  journal={arXiv preprint arXiv:2606.29667},
  year={2026},
  doi={https://doi.org/10.48550/arXiv.2606.29667}
}