Creative Adversarial Networks (CAN) - Generating Art by Learning About Styles and then Deviating from Style Norms

 What is art?  What is creativity?  What is 'aesthetically appealing?  These are fascinating questions that perhaps tell us more about what it is like to be human.

Which poses the following question.  Is it possible to remove the human completely from the art making process and still generate what humans would consider art?


Here's a link to a blog article on this topic.  It points out something interesting, which is that conventional GAN systems might be great at creating new images of a particular type (faces, landscapes, furniture), but in some sense they aren't really being creative.  At least not in a certain artistic sense.  Because they aren't breaking away from the representations they were trained to generate.

The developers of these GAN systems would probably reply, of course not, because it would defeat the whole point of our trained GAN system, which is to generate things that look like the type of images they were trained on.

But in some sense, the whole point of being creative is to break away from convention.  To create something new, different.  To explore the creative space of possibilities.



Here's a link to a technical paper on this topic called 'CAN: Creative Adversarial Networks, Generating "Art" by Learning About Styles and Deviating from Style Norms'.  It introduce a new kind of GAN, called a Creative Adversarial Network (CAN).  

Here's an example of some CAN system generated artwork from the paper linked above.

A CAN generator starts with random noise like a GAN, but receives 2 different signals from the CAN system's discriminator.  The first is a classification of 'art or not art'.  The second is a 'style ambiguity' signal that measures how confused the discriminator is in trying to identify the output of the generator as art of known styles.  This allows a CAN system to train it's generator to output images the discriminator recognizes as 'art', but that have a level of style ambiguity.

Here's a block diagram of the CAN architecture from the paper linked above.


The notion of having a discriminator that computes multiple outputs, and then passing that back to a GAN generator is interesting.  This work was done at the The Art and Artificial Intelligence Laboratory at Rutgers University.


Comments

  1. SynthetikNovember 28, 2020 at 10:04 AM

    So i looked at another related article on above that describes an art installation in NYC based on this algorithm. And it showed off a bunch of different images generated by the algorithm as a part of the art gallery installation.

    I have to be honest, the imagery in the art gallery installation looked a lot like MSG (modular synthesized graphics) procedural imagery that one could have generated in Studio Artist 10 years ago or more. Note that the imagery could be completely generated by the MSG system in Studio Artist without any human intervention other than grabbing ones they liked (and i'm sure they did exactly the same for the Ai art exhibit i'm referring to, ie: hand picked the 'best' output images for their gallery exhibition).

    So i think we need to be a little bit cautious about rave reviews of 'neural net generated art' or AI art, where reviewers or the media get caught up in the hype of AI Art.

    ReplyDelete

Post a Comment

Popular posts from this blog

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

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

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