LayoutPrompter is a versatile method for graphic layout generation, capable of solving various conditional layout generation tasks (as illustrated on the left side) across a range of layout domains (as illustrated on the right side) without any model training or fine-tuning.
We conduct experiments on three groups of layout generation tasks, including
- constraint-explicit layout generation
- content-aware layout generation
- text-to-layout
Below are the qualitative results.
- Clone this repository
git clone https://github.com/microsoft/LayoutGeneration.git
cd LayoutGeneration/LayoutPrompter
- Create a conda environment
conda create -n layoutprompter python=3.8
conda activate layoutprompter
- Install PyTorch and other dependencies
conda install pytorch=1.13.1 torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
pip install -r requirements.txt
pip install -e src/
We use 4 datasets in this work, including RICO, PubLayNet, PosterLayout and WebUI.
They can be downloaded from HuggingFace using the following commands:
git lfs install
git clone https://huggingface.co/datasets/KyleLin/LayoutPrompter
Move the contents to the dataset directory as follows:
dataset/
├── posterlayout
├── publaynet
├── rico
├── webui
We include three jupyter notebooks here, each corresponding to a type of layout generation task. They all consist of the following components:
- Configuration
- Process raw data
- Dynamic exemplar selection
- Input-output serialization
- Call GPT
- Parsing
- Layout ranking
- Visualization
Try it!
If you find this code useful for your research, please cite our paper:
@inproceedings{lin2023layoutprompter,
title={LayoutPrompter: Awaken the Design Ability of Large Language Models},
author={Lin, Jiawei and Guo, Jiaqi and Sun, Shizhao and Yang, Zijiang James and Lou, Jian-Guang and Zhang, Dongmei},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023}
}