Using AI and the Style GAN variant, we were able to train a dataset of ~35,000 Pokemon images, we were able to create a few new suggested Pokemon.
Pokemon was first created in 1996. In the twenty years, it has become one of the most recognizable franchises in the world. By March 2020:
Pokémon video games have sold over 380 million copies worldwide.
Pokémon Go is the most-downloaded mobile game, with over 1 billion downloads.
Pokémon Trading Cards have sold over 34.1 billion cards.
The entire media franchise is the highest-grossing entertainment media franchise of all time, grossing an estimated $90 billion in lifetime revenue
Yet, after all that, there are still
only a total of 893 unique Pokemon
With AI, we can change that.
An intuitive understanding of GANs, Progressive GAN’s and Style GANs
GAN’s first hit the stage in 2014, and garnered a lot of hype by generating images that looked reasonably realistic, was easy to train and sample from, and was theoretically satisfying. I have aseparate post
discussing the GAN formulation in more depth.
From the original generative adversarial network paper
Since then, a series of changes have been made to the basic GAN to generate some amazingly realistic looking images.
GAN improvements through the ages
Traditional Generative Adversarial Network Architecture
is an attempt to fix a fundamental problem with traditional GAN’s which is that the generator has a much harder job than the discriminator, especially with high resolution images. (In the same way it’s easier for you to classify an image as being a cat or dog than it would be to draw an image of a cat or dog.)
Progressive GAN’s alter the training phase of traditional GAN’s by first training a single layer GAN against a 4x4 image, then a two layer GAN against an 8x8 image, reusing the already trained 4x4 layer. This makes the problem for the generator easier during the earlier phases of training, and scales up towards generating high resolution images.
Progressive GAN’s start by learning how to generate low resolution images and scale up as training progresses.
However, both traditional GAN’s and Progressive GAN’s offer little control over the image produced by the generator, for instance, generating a person with dark hair vs light hair. Or female vs male. Style GAN’s extend the architecture of traditional GAN’s to allow for more control of the image produced by the generator.
The style vectors that are used can then be reused from one generated image to the next, by injecting the same style vector, thereby “transferring” the style (hair color, age, gender, etc) from one image to another.
Style transfer at different layers in the generator
How do I build one?
There’s a bit too much code to a Style GAN to be able to review all of it, but we can go through each important piece with some code samples.
Building off Progressive GAN’s
Addition of mapping network
Adaptive instance normalization
Removal of latent vector input to generator
Addition of noise to each block
What a Style GAN adds to a Progressive GAN
The Style GAN starts from Progressive GAN’s and adds a series of improvements.
Here is the StyleGAN architecture. Notice that the latent variable
is not directly passed into the generator. Rather, it’s passed through an 8 layer MLP, called the mapping network, to generate a “style vector”.
This style vector is then injected into every layer of the generator network (which is referred to as the synthesis network in this paper). The style vector is used to produce a scale and bias vector for each channel in the image as part of the adaptive instance normalization.
1 2 3 4 5 6 7 8 9 10flatten = lambda t: [item for sublist in t for item in sublist] class MappingNetwork(nn.Module): def __init__(self): super().__init__() blocks = [[nn.Linear(512, 512), nn.LeakyReLU(0.2)] for _ in range(8)] self.net = nn.Sequential(*flatten(blocks)) def forward(self, latent_vector): style_vector = self.net(latent_vector) return style_vector
Adaptive Instance Normalization (AdaIN)
To understand how the adaptive instance normalization works, let’s first review how batch normalization works.
Standard batch normalization formulation
Adaptive. instance normalization formulation
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25class AdaIN(nn.Module): def __init__(self): super().__init__() def mu(self, x): """ Takes a (n,c,h,w) tensor as input and returns the average across it's spatial dimensions as (h,w) tensor [See eq. 5 of paper]""" n = torch.sum(x,(2,3)) d = x.shape*x.shape return n / d def sigma(self, x): """ Takes a (n,c,h,w) tensor as input and returns the standard deviation across it's spatial dimensions as (h,w) tensor [See eq. 6 of paper] Note the permutations are required for broadcasting""" n = torch.sum((x.permute([2,3,0,1])-self.mu(x)).permute([2,3,0,1])**2,(2,3)) d = x.shape*x.shape return torch.sqrt(n / d) def forward(self, x, y): """ Takes a content embeding x and a style embeding y and changes transforms the mean and standard deviation of the content embedding to that of the style. [See eq. 8 of paper] Note the permutations are required for broadcasting""" return (self.sigma(y)*((x.permute([2,3,0,1])-self.mu(x))/self.sigma(x)) + self.mu(y)).permute([2,3,0,1])
Addition of Gaussian Noise
Gaussian noise is added to the activation maps before each AdaIN operation which is used to generate style variation at each level of the generator.
The added noise is what allows for a generated image to be identical overall but still output variation at local levels
(a) Noise is applied to all layers. (b) No noise. Noise in fine layers only (64–1024 ). (d) Noise in coarse layers only (4–32 ).
Removal of Latent Vector Input to Generator
Traditional GAN’s generate a latent vector, which is fed to the generator, to produce samples. Style GAN’s actually just use a constant 4x4x512 vector that is kept the same during training and inference. This is possible because the variant comes from the style vectors and gaussian noise added at each layer via the AdaIN operation, rather than the initial latent vector.
Remember that the whole point of Style GAN’s was to discover style vectors that could be reapplied from one image to another. This only works if both images start from the same “canvas”, which is what this 4x4x512 constant acts as. If the canvas changes, then the style vectors learned would not be transferable.
Training on a Pokemon Dataset
It took about a week to train a model to produce results that looked reasonable on an Nvidia GeForce 3070 GPU. I’m unsure how much better the images would look after a few more weeks of training, but I’ll keep it running for a bit longer.
A Few Suggestions for New Pokemon