SCARLET (Scientific Chart and Figure Reasoning Dataset) is a large-scale multimodal reasoning dataset constructed from figures extracted from publicly available arXiv research papers.
The dataset is designed for training and evaluating Vision-Language Models (VLMs) on scientific figure understanding and visual reasoning.
Each figure is paired with multiple question-answer instances spanning several reasoning categories. Every QA pair additionally contains an… See the full description on the dataset page:
https://huggingface.co/datasets/AarIsJay/SCARLET.
