FPD is the first manually annotated dataset that focuses on fine-grained segmentation, which consists of 66 non-Manhattan layout documents (shown in Fig.~\ref{fig:fpd}). In the labeling process, we use point labeling and polygon labeling, and the label export style is ``cvat for images 1.1". The data comes from the pages of complex Chinese and English magazines, and the page size is not fixed.
For more information, please consult our paper here.
If you would like to download the FPD dataset, please send email to [email protected] and [email protected]. We will send you the dataset download link.
If you use the FPD data please cite:
@inproceedings{ma2023image,
title={Image Layer Modeling for Complex Document Layout Generation},
author={Ma, Tianlong and Wu, Xingjiao and Du, Xiangcheng and Wang, Yanlong and Jin, Cheng},
booktitle={2023 IEEE International Conference on Multimedia and Expo (ICME)},
pages={2261--2266},
year={2023},
organization={IEEE}
}