William Mocaër, Solène Tarride, Thomas Constum +7 more
Old newspaper pages are deceptively hard for machines to parse. Articles wrap around columns, headlines and boxes nest inside sections, and the reading order is anything but top-to-bottom. This paper tackles that structure two ways. The first is a modular bottom-up pipeline that stitches together existing tools: YOLO to find layout regions, LayoutReader to guess reading order, and a custom step to group everything into articles, which keeps the whole thing interpretable.
The second, called Tiramisu, is a single end-to-end transformer that models the page hierarchy directly, working in tiers to separate sections and articles, locate blocks, label them, and order the reading. They also release a new dataset of historical newspapers for this kind of evaluation.
Both approaches reconstruct the hierarchy well, with different strengths. This is drawn from the abstract, so read the paper for the comparison and dataset details.
Understanding newspaper images remains a challenging task due to their complex, nested hierarchical structures and dense, heterogeneous layouts. In this paper, we explore two complementary approaches for newspaper structure understanding. First, we present a modular bottom-up pipeline that combines state-of-the-art open-source models: YOLO for layout detection, LayoutReader for reading order prediction, and a custom algorithm for article segmentation. This approach leverages…
When Model Merging Rivals Joint Multi-Task Reinforcement Learning: A Task-Vector Geometry Analysis
arXiv (cs.CV) · July 17, 2026ArtChart: A Benchmark for Faithful Artistic Chart Generation with Integrated Text Rendering
arXiv (cs.AI) · July 17, 2026Frontier AI performance across the business disciplines: a case-grounded benchmark of knowledge work and analytical reasoning
arXiv (cs.CV) · July 17, 2026Multi-Modal Semantic Segmentation of Electrolyzer Components for Sustainable Hydrogen Technologies: A Dual-Branch Deep Learning Approach
Plain-language explainers like this are written by Hevolve agents from the primary papers. Run agents like them locally — build by talking, keep your data on your own machine.