image_to_image_translation_cyclegan's Introduction

image_to_image_translation_cyclegan

Setup

Download the Monet2Photos dataset as follows:

wget https://people.eecs.berkeley.edu/~taesung_park/CycleGAN/datasets/monet2photo.zip

Make a new directory (e.g. 'data'), and make it the current directory

mkdir data
cd data

Unzip the zip file containing the dataset inside the 'data' directory

unzip monet2photo.zip

Install the keras_contrib library as follows:

pip install git+https://www.github.com/keras-team/keras-contrib.git

Create a new directory (e.g. 'results') to store the original, generated and reconstructed images

mkdir results

About the execution modes

A) Train mode

Vanilla execution mode
Images will be loaded, and the generator and discriminator networks will be trained
Generator will generate images from the given input, discriminator will distinguish between real and fake images
The adversarial model is responsible for reaching an optimum value of the objective function
After every epoch, the weights are updated to tune both the networks to reach maximum efficiency.

B) Predict mode

The discriminator network isn't involved here - only the weights of the generator network are loaded, and the generator network is trained
The generator generates images for the input images it is provided
No networks are optimized, the generator just generates images for the entire batch of inputs.

image_to_image_translation_cyclegan's People

Contributors

Stargazers

Watchers

image_to_image_translation_cyclegan's Issues

write_log function no longer working in Tensorflow 2?

I have to comment off the lines for the time being:

     # write_log(tensorboard, 'discriminator_loss', np.mean(D_losses), epoch)
     # write_log(tensorboard, 'generator_loss', np.mean(G_losses), epoch)

Typo in line 440

line 440

    for i in range(len(generatedA)):
      save_images(originalA = batchA[i], generatedB = generatedB[i], reconstructedA = recons_A[i],
                  originalB = batchB[i], generatedA = generatedA[i], reconstructedB = recons_B[i],
                  path = "results/test_{}".format(i))

recons_A should be reconsA
recons_B should be reconsB

Recommend Projects