Skip to content

ubc-vision/attribute-guided-image-generation-from-layout

Repository files navigation

attribute-guided-image-generation-from-layout

This is the code repository complementing the paper attribute-guided image generation from layout. The pretrained model can be downloaded here: (TODO)

Model Pipeline

Dependencies and Setup

  • Python 3.6.8
  • SPADE
  • numpy==1.16.3
  • torch==1.1.0
  • torchvision==0.3.0
  • PIL==6.0.0
  • imageio==2.5.0
  • tensorboardX==1.8
  • sklearn==0.20.3
  • h5py==2.8.0

Experiments

Datasets

Visual Genome (VG) Download:

cd data/Dataset/vg
chmod +x download_vg.sh
./download_vg.sh

Preprocessing:

cd data/Dataset/vg
python train_test_split.py # split the dataset
cd ../..
python preprocess_vg.py

Train

To train 64x64 (or 128x128) from scratch, run

python train64.py

The script will check if pretrained models exist.

Training 128x128 model is computationally expensive. You might consider using torch.DataParallel for training with a larger batch size.

Test

To test 64x64 (or 128x128) model, run

python test64.py

You can modify attributes as you wish in the test64.py (test128.py) script.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages