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:
- PyTorch, version 1.2 (Although newer versions are also posible to use)
- matplotlib, tqdm, numpy and scipy
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):