PubLayNet

PubLayNet is a large-scale dataset designed for document layout analysis. It comprises over 330,000 scientific paper documents. The primary goal of PubLayNet is to facilitate research in automatically understanding the structural organization of scientific articles, such as identifying different regions like titles, paragraphs, figures, tables, and lists.

The dataset was created by extracting document images from PDF files found in the PubMed Central Open Access Subset. Each document in PubLayNet is annotated with bounding boxes that delineate various layout elements. This detailed annotation allows researchers to train and evaluate machine learning models for tasks like document segmentation and information extraction.

PubLayNet is particularly useful for developing models that can parse complex documents, which is a crucial step in digitizing and analyzing large volumes of scientific literature. Its scale and the diversity of document structures it represents make it a valuable resource for advancing the field of document image analysis and understanding. The dataset is publicly available, promoting open research and collaboration.