/ Machine Learning

Jokeriser with CycleGAN

Jokeriser with CycleGAN

jokerise

Not sure if Joker face would look good on you for Halloween? Try jokeriser!

Jokeriser finds your face with facenet_pytorch and translate your face to a Joker's using a generator trained with CycleGAN.

Getting Started

$ git clone https://github.com/junkwhinger/jokerise.git
$ cd jokeriser

Install dependencies

pip install -r requirements.txt

How to jokerise

# image
python demo.py --input samples/joaquin.jpg

# video
python demo.py --input samples/joaquin.mp4

# webcam
python demo.py --webcam

# wanna see original & tranlsated version side by side?
python demo.py --webcam --show-original

Note

  • I tested my code in OSX environment. Not sure if it works well on other platforms.

CycleGAN Training Details

  • Dataset
    • Joaquin Phoenix's and Heath Ledger's joker faces from Google (300 images)
    • Randomly selected faces from CelebA dataset (300 images)
  • Preprocessing
  • Training details
    • image size: 3x128x128
    • number of residual blocks: 6
    • epochs: 200

Have Fun!

Jokeriser-with-CycleGANv

GitHub