Nike Shoes generation using Deep convolutional Generative Adversarial Networks example.

The Deep Convolutional Generative Adversary Network(DCGAN) is based on the idea of Generative Adversary Network(GAN) first used at University of Montreal by Ian J. Goodfellow and his colleaques. Here is the original paper for GANs published in 2014. This repo presents an example of using DCGAN to generate Nike shoes from 100 Nike shoes downloaded from the internet.

Original Images used

File Organisation

• functions.py - This file contains all the functions used in this repo.
• app.ipynb - This is the Ipython notebook originally used to train the network.
• app.ppy - This is the python file which is an alternative to the Ipython Notebook that can be ran from the command line.
• Shoes - Folder containing shoes used as training data for this model.

Usage

• This project can be ran directly from the notebook and this line input_images_path =glob("Shoes/*") changed to the path images that will be used to train the network. The generator generates images of size (128x128x3) which is also the input to the discriminator. In order to change the image size, the generator has to be adapted to output the image size and the image_size = (128,128,3) in the second cell of the Notebook changed to the corresponding size. Please, endevour to use a squared iamge size to improve performance of model.
• The model can also be ran from the command line. Just type python app.py -h to get help on how to run model:

python app.py -h

Examples

• python app.py Shoes/

  • Run model on images in path Shoes/

• python app.py Shoes/ --b 20 --e 100 --s 5 --d 20

  • Path to images Shoes/.
  • Batch size: 20
  • Number of Epochs: 100
  • Show loss after a count of 5
  • Display generated images after a count of 20

Results

These are the results obtained using just 100 downloaded Nike shoes. These shoes are included in the repo here.

License

Anybody is free to download and make changes to this project as long as they reference this work.