Adversarial autoencoder TF2

A Tensorflow 2.0 implementation of Adversarial Autoencoder (ICLR 2016)

Model

Architecture	Description
	Regularization of the hidden code by incorporationg full label information (Fig.3 from the paper). _{Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, and Ian J. Goodfellow. 2015. Adversarial Autoencoders. CoRRabs/1511.05644 (2015). Figure 3 from the paper.}

Target prior distribution	Learnt latent space	Sampled decoder ouput

Input images	Reconstructed images

Gan	Encoder	Discriminator

python train_model.py --prior_type gaussian_mixture

--prior_type: Type of target prior distribution. Default: gaussian_mixture. Required.
--results_dir: Training visualization directory. Default: results. Created if non-existent.
--log_dir: Log directory (Tensorboard). Default: logs. Created if non-existent.
--gm_x_stddev: Gaussian mixture prior: standard deviation for the x coord. Default: 0.5
--gm_y_stddev: Gaussian mixture prior: standard deviation for the y coord. Default: 0.1
--n_epochs: Number of epochs. Default: 20
--learning_rate: Learning rate. Default: 0.001
--batch_size: Batch size. Default: 128
--num_classes: Number of classes (for further use). Default: 10

Visualization of outliers from learnt distribution in the latent space