Semantic Image Synthesis with Spatially Adaptive Normalization

 A lot of work has been done developing deep learning nets that take an image as input and output a series of tags for objects present in the image.  Less studied, but perhaps more interesting, is the notion of a deep learning net that takes a textural description of an image or scene and then outputs an image that looks like the textural description passed to it as input.

So let's deep dive into a recent paper that tries to turn a textural description of some imaginary image, and then generate an artificial image that looks like the description. And this particular approach adds an additional constraint, a segmentation map constraint to be precise.  And it being precise is kind of the whole point to this.

The paper describing this 'Semantic image Synthesis with Spatially-Adaptive Normalization' can be found here.

A project page called 'Semantic Image Synthesis with SPADE' can be found here.


Here's a short video presentation of 'GauGAN: Semantic Image Synthesis with Spatially Adaptive Normalization'.


Divyansh Jha has put together a really great blog post on Implementing SPADE using Fastai.  Perfect for HTC fastai api oriented deep learning folks.  Check it out.


And of course the world moves on, and people are trying to improve SPADE performance.  A recent attempt at improving SPADE is the Semantic Region-adaptive Normalization (SEAN) algorithm.

SEAN is conditioned on segmentation masks that describe the semantic regions in the desired output image. Using SEAN normalization, a network architecture can be made to control the style of each semantic region individually.

The SEAN generator network is built on top of SPADE and contains three convolutional network layers with their biases and scales modulated separately by individual SEAN blocks. There are two inputs per SEAN block: the set of style codes for specific regions, and a semantic mask that defines regions to apply the style code.

The paper 'SEAN: Image Synthesis with Semantic Region-Adaptive Normalization' is available here.

And here's a short 5 minute paper presentation of 'Sean: Image Synthesis with Semantic Region-Adaptive Normalization'.

I find it fascinating that you can pay serious money for well put together courses on this stuff, but then they don't get around to actually covering the most recent developments, which is kind of the whole point of taking a course on something like GAN Image Synthesis in 2020. 

Fortunately, we are covering all of the recent developments here at HTC.  Like what we just covered above, an in Pix2Pix last week.  And our course is free. And if we miss anything you think we should be shouting about, let us know and we'll get something posted on it as well.


Comments

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