Re-implement CycleGAN in Tensorlayer
Switch branches/tags
Nothing to show
Clone or download
Latest commit 38faad7 Oct 9, 2017
Permalink
Failed to load latest commit information.
datasets/sunflower2daisy add sunflower2daisy Oct 3, 2017
figures add improvement Sep 27, 2017
tensorlayer working Sep 30, 2017
LICENSE Initial commit Sep 22, 2017
README.md working Sep 30, 2017
dataset_clean.py first commit Sep 22, 2017
dataset_crop.py first commit Sep 22, 2017
main.py correct feed Oct 9, 2017
model_deconv.py correct patch Sep 28, 2017
model_upsampling.py working Sep 30, 2017

README.md

CycleGAN_Tensorlayer

Re-implement CycleGAN in TensorLayer

  • Original CycleGAN
  • Improved CycleGAN with resize-convolution

Prerequisites:

  • TensorLayer
  • TensorFlow
  • Python

Run:

CUDA_VISIBLE_DEVICES=0 python main.py 

(if datasets are collected by yourself, you can use dataset_clean.py or dataset_crop.py to pre-process images)

Theory:

The generator process:

Image text

The discriminator process:

Image text

Result Improvement

  • Data augmentation
  • Resize convolution[4]
  • Instance normalization[5]

data augmentation:

Image text

Instance normalization(comparision by original paper https://arxiv.org/abs/1607.08022):

Image text

Resize convolution (Remove Checkerboard Artifacts):

Image text

Image text

Final Results:

Image text

Image text

Reference: