Fortnite - PUBG

Originally inspired from this blog

Converting Fortnite to PUBG (output: 256 * 256) using CycleGAN and vice versa using Keras

Demo

This Model was trained with 50,000 image each of Fortnite and PUBG
Results after 3 Epochs

Video link

Prerequisites

• python 3.5
• keras==2.1.5
• keras-contrib==2.0.8
• tensorflow==1.5.0
• opencv-python==3.4.0.12
• NVIDIA GPU CUDA

Get Started

1. Prepare Data:

• Copy your two directory containing images(Fortnite, PUBG) in 'data/raw/'
• Then run the below command. This will save the data in batch numpy format at 'data/prepared/'

# prepare data
python3 prepare_data.py

2. Build Model

• To build Model from scratch you can directly run 'fortnite_pubg.ipynb'
  OR
  
• Download checkpoint and move it to 'checkpoint/' and run 'fortnite_pubg.ipynb'

References

• NIPS2016 hacks
• eriklindernoren
• Image-to-Image Translation with Conditional Adversarial Networks