Dataset

SciIR-82k

Contains 82k+ scientific figures/subfigures with aligned reasoning annotations, Sci-RCoT captions, prompts, and article metadata for generation tasks.

Hugging Face ↗Jun 24, 2026

  • multimodal
  • text-to-image
  • scientific-figures
  • reasoning-annotations

What it is

82k+ scientific figure-text pairs with structured reasoning annotations for training and evaluating scientifically faithful image generation. 82k+ processed scientific figures or subfigures paired with 3-part reasoning structure, detailed Sci-RCoT captions, concise prompts, and full source provenance from CC BY 4.0 papers.

Key results

  • 82k+ image-text pairs
  • 3 reasoning dimensions: structure, process, law
  • CC BY 4.0 sources from Nature and Nature Communications

Useful for training and evaluating reasoning-aware scientific text-to-image and figure synthesis models.

One useful ML email a week.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.