Official implemenation of MPCC: Matching Priors and conditionals for Clustering. This respository is strongly based on ''The author's officially unofficial PyTorch BigGAN'' implementation.
You will need:
Note that the official FID score and IS are based on tensorflow implementations. You will need tensorflow 1.1 and 1.3 respectively to obtain these official metrics using inception_tf13_p.py
and fid_p.py
. You can find C10 inception metrics in here.
A jupyter notebook is provided to perform generation, reconstructions and predictions with MPCC. Additionally Cifar10 models weights are included.
Here are some samples of the generative model:
Cifar10 samples (every two columns a different cluster):
Cifar20 samples (every row a different cluster):
Omniglot samples (every row a different cluster):