control-VAE

This repository contains reproduction of main results that displayed in [arxiv] as a part of ML course.

ControlVAE: Tuning, Analytical Properties, and Performance Analysis

by Huajie Shao∗, Zhisheng Xiao∗, Shuochao Yao, Dachun Sun, Aston Zhang, Shengzhong Liu, Tarek Abdelzaher, Fellow, ACM

Datasets

MNIST

No need to upload data files

celebA

Before running the code, download .zip file from here [celebA].

dSprites

Before running the code, download full datased downloaded from the original repo. [dSrites].
Subset of 2000 random dSprites samples downloaded from the Neuromatch Academy tutorial [Colab tutorial].

Store files as described below in data folder.
Run main.py files to train the models.

        data
            ↳ celebA
                ↳ img_align_celeba.zip
            ↳ dSprites
                ↳ dsprites_ndarray_co1sh3sc6or40x32y32_64x64.npz
                ↳ dsprites_subset.npz
        Disentangling
            ↳ main.py
        Image Generation 
            ↳ celebA
                ↳ main.py
            ↳ MNIST
                ↳ main.py