https://github.com/parisneo/pokigen

https://github.com/parisneo/pokigen

Last synced: 21 days ago
JSON representation

• Host: GitHub
• URL: https://github.com/parisneo/pokigen
• Owner: ParisNeo
• License: mit
• Created: 2020-10-24T08:55:47.000Z (about 4 years ago)
• Default Branch: main
• Last Pushed: 2020-11-01T01:34:44.000Z (about 4 years ago)
• Last Synced: 2024-10-31T11:29:14.400Z (2 months ago)
• Language: Jupyter Notebook
• Size: 1.2 MB
• Stars: 0
• Watchers: 3
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Changelog: Changelog.txt
  • License: LICENSE

Awesome Lists containing this project

README

        
# Pokigen

## Description

A simple CNN Beta-VAE based Pokémon image generator written in simplified Tensorflow 2.3 Keras.

The CNN beta-VAE structure features a 

- Variational Convolution based Encoder encoding the images in a latent space as a Gaussian distribution consisting mainly in encoding blocks with different parameters depending on the version of the network you are using.

- Convolutional based decoder with upscaling decoding images from latent space back to normale images.

We present 3 varients of the network :

  - Light version : a simple CNN maxpooling, cnn upsampling architecture

  - Mobilenet : uses the convolutional layers of mobile net in encoder and cnn upsampling layers for decoding

  - Full : a fully fledged CNN encoder decoder with deep resnet structure. Needs alot of time and power to learn

The input of the autoencoder is a 128X128X3 image

The output is also a 128X128X3 image

# Objective

As for today, the software is a simple beta vae based pokemon images generator. The real objective is to correlate the pokemon name with its photo and to be able to generate pokemons from the name.