Learning Discrete and Continuous Factors of Data via Alternating Disentanglement

Demo

Dependency

• python=3.5
• tensorflow version = 1.4
• CUDA 8.0
• cuDNN 6.0
• Environment detail is listed in `ex.yml'

Citing this work

@inproceedings{jeongICML19,
    title={
        Learning Discrete and Continuous Factors of Data via Alternating Disentanglement
    },
    author= {Yeonwoo Jeong and Hyun Oh Song},
    booktitle={International Conference on Machine Learning (ICML)},
    year={2019}
}

Dataset(dSprites)

• Download from https://github.com/deepmind/dsprites-dataset

Edit path

• Edit path in 'config/path.py'
• ROOT - (directory for experiment result)
• DSPRITESPATH - (directory for downloaed dsprites)

Run model

• Dsprites_exp/CascadeVAE/main.py
• Dsprites_exp/CascadeVAE-C/main.py

Trained model

• Download from here.
• Here are trained models from 10 different random seeds.

License

MIT License