Gaussian Mixture VAE with Contrastive Learning for Multi-Label Classification (C-GMVAE)

The implementation of C-GMVAE using PyTorch.

Sample Dataset

We use mirflickr as our running example since it is commonly used and has a moderate size. Dataset location: data/mirflickr

Dependencies

• Python 3.7+
• PyTorch 1.7.0
• numpy 1.17.3
• sklearn 0.22.1

Older versions might work as well.

Run

To train the model: bash script/run_train_mirflickr.sh

To test the model (this .sh will be produced automatically): bash script/run_test_mirflickr.sh

The seed is 1 by default, but can be changed in the bash file.

Paper

If you find our work interesting, please consider citing the following paper:

@inproceedings{bai2022gaussian,
  title={Gaussian Mixture Variational Autoencoder with Contrastive Learning for Multi-Label Classification},
  author={Bai, Junwen and Kong, Shufeng and Gomes, Carla P},
  booktitle={International Conference on Machine Learning},
  pages={1383--1398},
  year={2022},
  organization={PMLR}
}

Related Repo

MPVAE