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.

Results for gaussian_mixture prior

Latent space

Target prior distribution	Learnt latent space	Sampled decoder ouput

Reconstruction

Input images	Reconstructed images

Training loss

Gan	Encoder	Discriminator

Example of usage

python train_model.py --prior_type gaussian_mixture

Attributes

• --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

Visualization of outliers from learnt distribution in the latent space