Capturing label characteristics in VAEs

This is the repository of the final project done for the MVA Master’s course “Probabilistic Graph Models” (Prof. Latouche and Prof. Mattei). We extend and run additional experiments on the model developed in Capturing Label Chacteristics in VAEs ICLR 2021. More precisely, you will find:

• A tutorial to create multi-class labels (e.g., $Hair \in [Brown \space Hair, Blond \space Hair, Bald, Black \space Hair]$. The transformed data is already present in the data folder
• Tutorials to setup the training (tutorial_data_loader.ipynb) in a binary class or multi-class fashion
• Adapted code of CCVAE model for multi-class training
• Several notebooks and the report.pdf that analyzes and discusses the results

Original Repository

Pytorch repository for Capturing Label Chacteristics in VAEs ICLR 2021. We kindly ask that you cite our work if you plan to use this codebase:

@inproceedings{
    joy2021capturing,
    title={Capturing Label Characteristics in {\{}VAE{\}}s},
    author={Tom Joy and Sebastian Schmon and Philip Torr and Siddharth N and Tom Rainforth},
    booktitle={International Conference on Learning Representations},
    year={2021},
    url={https://openreview.net/forum?id=wQRlSUZ5V7B}
}

### Usage

Ensure that CelebA is in the directory data/datasets/celeba, such that the path data/datasets/celeba/celeba/img_align_celeba/* is accessable.

To train, run:

python ss_vae.py -sup <sup-frac> --cuda>

where <sup-frac> is the fraction of supervised data (e.g. 0.004, 0.06, 0.2, 1.0).