GAN_cat_image_generator

Presentation

This Python project uses the PyTorch library to generate a unique cat image using GAN technology.

Installation

• Install Python libraries

pip install -r requirements.txt

When the program is run locally, the following files are downloaded locally into the cloned directory :

• dataset.zip
• dataset

They represent training and test data. They don't weigh much more than 50MB each. They can be easily deleted after training.

However, I recommend running this code in Google Colab to benefit from the computing power of their CPU and/or GPU. What's more, no files will be downloaded locally if the code is run on Google Colab.

Algorithm configuration

The internal values used for the algorithm can be modified in the TP_GAN_cat_image_generator file. To name just a few important ones:

BATCH_SIZE = 64
EPOCHS = 20
lr=2e-4
LATENT_DIM = 100

argument	type	description
BATCH_SIZE	int	The size of the data batch used for training
EPOCHS	int	The number of epochs used to train the model
lr	float	The learning rate of the optimizer (“Adam” here)
lr	int	The size of the latent vector from which the generator starts an image generation

Notes: All values must be modified with full knowledge of the facts. No check is made on the consistency of values. Execution time can be quite long (less than an hour on Colab for several weeks of epochs).

Examples

Here are some images of cats generated with 20 EPOCHS: