TakaoNarikawa/LayoutGAN-Reimplementation

PyTorch reimplementation of "LayoutGAN: Generating Graphic Layouts with Wireframe Discriminators" publishsed in ICLR 2019

LayoutGAN Reimplementation

PyTorch reimplementation of "LayoutGAN: Generating Graphic Layouts with Wireframe Discriminators" publishsed in ICLR 2019: https://openreview.net/forum?id=HJxB5sRcFQ.

Requirement

• PyTorch : 1.8.1
• TorchVision : 0.9.1
• PyTorch-Lightning : 1.3.4

Getting Started

Point Layout (using MNIST dataset)

1. Download pre_data_cls.npy from Link. This is from Official Tensorflow Implementation Repository
2. Run python3 train.py. Use --gpus 1 option for GPU.

BBox Layout (using PubLayNet dataset)

1. Download labels.tar.gz from PubLayNet Official and decompress it as below.

PubLayNet
├ train.json
├ val.json
└ preprocess.py

2. Run python3 preprocess.py in /PubLayNet. Then you will have train.npy and val.npy
3. Run python3 train.py --train_mode bbox. Use --gpus 1 option for GPU.

Results

Point Layout

BBox Layout