CVAE based School Idol image generation. Published in proc. of SSCC 2nd, 2017.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
data
.gitignore
LICENSE
README.md
model.py
preprocess.py
sample.py
train.py
utils.py

README.md

Love2Live

Zenkai No? Love Live!

Conditional Variational Autoencoder [1] based School Idol image generation, implemented in PyTorch.

Used 1,466 face images of members of main school idol teams on the animation "Love Live!", extracted from the game "Love Live! School Idol Festival".

Note that this repo's implementation is also a fine implementation of the CVAE.

Requirements

  • numpy
  • scipy
  • pytorch

Usage

Clone the repo and type following:

python preprocess.py
python train.py
python sample.py

Using a nice GPU is heavily recommended.

Results

Autoencoding images in a train set

train

Autoencoding images in a test set

test

Generating images from a randomly sampled latent vector

random

Interpolating conditions

Hanayo to Rin:

hanayo-to-rin

Maki to Nico:

maki-to-nico

Maki to Yoshiko:

maki-to-yoshiko

Ruby to Dia:

ruby-to-dia

Yoshiko to Riko:

yoshiko-to-riko

[1] Kingma, Diederik P., et al. "Semi-supervised learning with deep generative models." Advances in Neural Information Processing Systems. 2014.