mini-project to Deep Learning School
In this assignment, the implementation of CycleGAN was chosen using the example of jeans. The task was to turn black jeans into blue and blue into black. For this task, a dataset was selected with a large number of clothes and markup of categories and colors.
So I got a dataset of 800 black and 800 blue jeans. They were divided into a training and test set as follows: 650 blue jeans for a training set, 150 blue jeans for a test set, 650 black jeans for a training set, 150 black jeans for a test set. Dataset there.
The model was trained for ~16 hours by Google Colab Pro. This version has 200 epochs. Weights there.
Some results are here:
Format: original image -> generated image, original image -> generated image
