PP-DocLayout-L

PP-DocLayout-L is the high-precision model in the PP-DocLayout series: it delivers 90.4% [email protected] for document layout detection.

This is your top-tier document layout detection model, built on the RT-DETR-L architecture and optimized by the PaddlePaddle Team (Baidu Inc.). PP-DocLayout-L achieves a verified 90.4% [email protected], localizing structural elements across diverse documents: think academic papers, financial reports, and contracts. It processes a full page in just 13.4 ms on a T4 GPU, efficiently recognizing 23 distinct layout categories (e.g., titles, tables, formulas, and figures). This performance profile makes it the definitive choice for high-accuracy, large-scale data construction and document intelligence pipelines.

https://github.com/PaddlePaddle/PaddleX