SPN2D-GAN in PyTorch

Paper: 《SPN2D-GAN: A Night-to-day Image-to-Image Translator Based on Semantic Prior Generator》

Usage

├── datasets
   ├── ACDC
       ├── rgb_anon_trainvaltest
           └── ...
       └── gt_trainval
           └── ...
   └── YOUR_DATASET_NAME
       ├── trainA
           ├── xxx.png
           ├── yyy.png
           └── ...
       ├── trainB
           ├── zzz.png
           ├── www.png
           └── ...
       ├── testA
           ├── aaa.png
           ├── bbb.png
           └── ...
       └── testB
           ├── ccc.png
           ├── ddd.png
           └── ...

• The ACDC-night train and val set images need to be placed in trainA and trainB, and ACDC-night test set images need to be placed in testA and testB

Prerequisites

• Linux Ubuntu + Anaconda
• Python 3.7
• PyTorch 1.5.0 + cuda 10.2 + cudnn 7.6.5 + torchvision 0.6.0
• dominate 2.6.0
• tensorboardX 2.2
• kornia 0.3.0

Train

• Train a model:

  > python train.py --dataroot ./datasets/YOUR_DATASET_NAME --name SPN2DGAN

• pretrained semantic mask extraction network and semantic supervision network model: Google_drive
• Note that the path to the ACDC dataset needs to be changed manually in data/unaligned_dataset.py.

Test

• Test the model:

  > python test.py --dataroot ./datasets/YOUR_DATASET_NAME --name SPN2DGAN

Metric

> python fid.py ./testResult/real_day ./testResult/fake_day --gpu 0

• The generated daytime images need to be placed in the /fake_day folder and the ACDC-test set real-world daytime images need to be placed in the /real_day folder.

Dataset

Download the ACDC-night dataset.

Acknowledgments

Our code is inspired by pytorch-CycleGAN.