PosterLayout: A New Benchmark and Approach for Content-aware Visual-Textual Presentation Layout

This repository contains the guidelines of benchmark PKU PosterLayout and Pytorch implementation of DS-GAN for "PosterLayout: A New Benchmark and Approach for Content-aware Visual-Textual Presentation Layout", CVPR 2023.

For dataset details and downloads, please visit our project page.

Comparison of layouts generated by different approaches.

How to Run

Prerequisites

If your operating system is linux-64, directly run

conda create --name yourenvname --file spec-file.txt

Otherwise, try

pip install -r requirements.txt

• Environment

Python 3.9
CUDA 11.0

• Module

torch==1.12.1
torchvision==0.13.1
timm==0.6.5
opencv-python==4.6.0.66
pandas==1.4.3
Pillow==9.2.0

Models

1. Download pre-trained weights from PKU Netdisk(pw: P04X) or Google Drive
2. Put corresponding .pth files under model_weight/ or output/, as follow:

model_weight/
├─ resnet18-5c106cde.pth
├─ resnet50_a1_0-14fe96d1.pth
output/
├─ DS-GAN-Epoch300.pth

Dataset

1. Download PKU PosterLayout from the project page
2. Unzip compressed files to corresponding directories
3. Put directories under Dataset/, as follow:

Dataset/
├─ train/
│  ├─ inpainted_poster/
│  ├─ saliencymaps_basnet/
│  ├─ saliencymaps_pfpn/
├─ test/
│  ├─ image_canvas/
│  ├─ saliencymaps_basnet/
│  ├─ saliencymaps_pfpn/
├─ train_csv_9973.csv

Usage

• Training

sh train.sh

• Testing and Evaluating

sh test_n_eval.sh

Citation

If our work is helpful for your research, please cite our paper:

@inproceedings{Hsu-2023-posterlayout,
    title={PosterLayout: A New Benchmark and  Approach for Content-Aware Visual-Textual Presentation Layout},
    author={HsiaoYuan Hsu, Xiangteng He, Yuxin Peng, Hao Kong and Qing Zhang},
    booktitle={Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    year={2023},
    pages={6018-6026}
}

Contact us

For any questions or further information, please email Ms. Hsu (kslh99@stu.pku.edu.cn).