Towards Hierarchical Structure Understanding of Newspaper Images
The paper pairs a modular newspaper-layout pipeline with a new transformer architecture, Tiramisu, for reconstructing nested article structure.
The authors frame newspaper understanding as a hierarchy problem: sections, articles, blocks, categories, and reading order. Their bottom-up system combines YOLO, LayoutReader, and a custom article segmentation algorithm. Tiramisu handles the same task end to end with tiered transformer stages and parallel attention. The paper also releases Finlam La Liberté, a dataset for evaluating hierarchical retrieval in historical newspapers. ArXiv · AI/CL/LG's note
The authors frame newspaper understanding as a hierarchy problem: sections, articles, blocks, categories, and reading order. Their bottom-up system combines YOLO, LayoutReader, and a custom article segmentation algorithm. Tiramisu handles the same task end to end with tiered transformer stages and parallel attention. The paper also releases Finlam La Liberté, a dataset for evaluating hierarchical retrieval in historical newspapers. ArXiv · AI/CL/LG's note
score 4