onestage_julia

Julia implementation of the CVPR 2021 paper Training Generative Adversarial Networks in One Stage

Baseline Model Updates

• src includes the layers and networks
• utils includes some utility functions
• main.ipynb includes the main training pipeline.
• trained_model include a generator and a discriminator checkpoint, trained on MNIST for 20 epochs for the two stage setup.

Current Status

Currently, I am able to train the two stage model on MNIST, and it produces good outputs. These outputs can be seen on main.ipynb. I will continue with the implementation of the one stage version soon.

TODO Items

• Batch Normalization layer gives a weird CUDA error, fix that. ✔
• Complete the training loop by implementing the optimizers. ✔
• Fix the bug in the training, so that the model is outputting meanningful results. ✔
• Complete the dataloader for CelebA too.
• Implement the onestage version.

@InProceedings{shen2021training,
    author    = {Shen, Chengchao and Yin, Youtan and Wang, Xinchao and Li, Xubin and Song, Jie and Song, Mingli},
    title     = {Training Generative Adversarial Networks in One Stage},
    booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {3350-3360}
}