Adversarial Latent Autoencoders

 Now you may have noticed that the auto-encoder architecture is having a resurgence recently as an alternative to the GAN architecture for constructing a generative model.   Take a look at yesterday's post for one example. 

Today's post continues the trend by taking a look at the Adversarial Latent Autoencoder (basic architecture shown below).

The ALAE architecture is a modification of original GAN by decomposing the Generator(G)(\textbf{G}) and Discriminator(D)(\textbf{D}) networks into two networks such that: G\textbf{G} = G∘F\textit{G} \circ \textit{F} and D\textbf{D} = D∘E\textit{D} \circ \textit{E}. The architecture is shown in Figure 3. It's assumed that the latent space between both the decomposed networks is same and is denoted as W\mathcal{W}.


Below is a modification of ALAE to get StyleALAE

There are two components of StyleALAE:

  1. The generator of ALAE is replaced with the generator of StyleGAN as shown in the right side of figure 4.
  2. The left side of figure 4 is a symmetrical Encoder so that style information can be extracted which drives the StyleGAN generator.

The style information is extracted from the ithi^{th} layer by introducing Instance Normalization(IN) in that layer. This layer outputs channel-wise averages (μ)(\mu) and standard deviation (σ)(\sigma), which represents the style content in each layer. The IN layer provides normalization to the input in each layer. The style content of each such layer of the encoder is used as input by the Adaptive Instance Normalization (AdaIN) layer of the symmetric generator which is linearly related to the latent space ω\omega. Thus, the style content of the encoder is mapped to the latent space via a multilinear map.

Here's a short video someone put together that showcases this work, contrasting it to StyleGAN.


Here's a link to the paper.

Here's a link to the official code depository for the paper.

There are a number of different implementations of this algorithm available in code here.  I'll pick a good PyTorch one to showcase at a later time (there are multiple ones at the link).

Here's a link to a blog article that discusses the architecture in detail (including the quotes i grabbed above for the description of the 2 figures).

Comments

Post a Comment

Popular posts from this blog

Pix2Pix: a GAN architecture for image to image transformation

Image
 I thought following up yesterday's TraVelGAN post with a Pix2Pix GAN post would be useful to compare what is going on in the 2 architectures.  Two different approaches to the same problem. I stole this Pix2Pix Overview slide below from an excellent deeplearning.ai GAN course (note that they borrowed it from the original paper) because it gives you a good feel for what is going on inside of the Pix2Pix architecture.    Note how the Generator part is very much like an auto-encoder architecture, but rebuilt using the U-Net architecture features (based on skip-connections) that fastai has been discussing in their courses for several years before it became more widely known to the deep learning community at large  (and which originally came from an obscure medical image segmentation paper) . So the Generator in this Pix2Pix GAN is really pretty sophisticated, consisting of a whole image to image auto-encoder network with U-Net skip connections to generate better image quality at highe
Read more

CycleGAN: a GAN architecture for learning unpaired image to image transformations

Image
 Like the Doublemint Twins touting the joys of Doublemint gum, 2 GANs are surely better than 1 GAN.  Especially if we package them together inside of one meta GAN module.  And this is exactly what the CycleGAN architecture does. Have you ever harbored dark secrets of turning a horse into a zebra?  The CycleGAN was developed to do just that. Learn how to turn a horse into a zebra.  And more. Now right away you can notice a difference between the image to image transformation GAN architectures we've been discussing over the last few posts.  Those last few posts described systems that learn from a database of matched input-output image pairs. And if your goal is to turn an edge representation into a nicely filled in continuous tone image, it's easy to build your database of matched input-output image pairs that your GAN system can then learn off of.  Take a continuous tone image (which will be the output of the database pair entry), then run it through an edge detector algorithm. 
Read more

Smart Fabrics

Image
"What makes smart fabrics revolutionary is that they have the ability to do many things that traditional fabrics cannot, including communicate, transform, conduct energy and even grow."  This is a quote from Rebeccah Pailes-Friedman, who is a professor at Pratt with a research focus on developing wearable technology and smart textiles. Enhancing textiles can be viewed through different lens.  Enhancements can be aesthetic, or performance boosting, or durability boosting. 
Read more