TODO.txt

TODO (VAES)
- (DONE 12/17) vae for mnist
- (DONE 03/17) refactor code
- (DONE 03/17) cvae for mnist
- **vae for cifar
- **dc-vae for cifar
- **dc-cvae for cifar

TODO (GANS)
- (DONE 06/17) gan for mnist
    - choose cost functions
    - train w/ labels/one-sided label smoothing?
- **cgan for mnist
- ***dc-gan for cifar
    - reference batch normalization/virtual batch normalization
- **dc-cgan for cifar

TODO (OTHER)
- (DONE 04/16) abstract data loading/management class similar to tf's mnist
- **get cifar data loader working
- consolidate vae and cvae into single class?
- plug and play
- visualization in open(web?)gl


Resources:
- Generative visual manipulation on the natural image manifold paper
  https://people.eecs.berkeley.edu/~junyanz/projects/gvm/eccv16_gvm.pdf
- DCGAN code
  https://github.com/Newmu/dcgan_code
- autoencoding beyond pixels using a learned similarity metric
  https://arxiv.org/pdf/1512.09300.pdf
- high resolution: CPPN w/ GAN+VAE
  http://blog.otoro.net/2016/04/01/generating-large-images-from-latent-vectors/