Generative Adversarial Networks for Image Synthesis and Transformation

 We're continuing our exploration of generative neural models with this presentation by Jan Kautz from Nvidia on 'Generative Adversarial Networks for Image Synthesis and Translation'.  This presentation is from November 2019.  It's a good overview of some of the different GAN approaches for generative synthesis we've recently been looking into.

I find the use of 'translation' to be confusing since it typically refers to an affine transformation, so the post is titled using our preferred 'Transformation' terminology.


We've got a GAN double header today.  Our second presentation is from DeepMind, and is a Deep Learning Lecture on the topic of 'Generative Adversarial Networks'.



Mihaela Rosca presents Part 1- 'Overview', and Part 2-' Evaluating GANs'.  Jeff Donahue then presents The GAN Zoo, which has a Part 3.1 on 'Image Synthesis with GANs: MNIST to ImageNet' , and a Part 3.2 on 'GANs for Representation Learning', and a Part 3.3 on 'GANs for Other Modalities and Problems'.

I found the second half of this presentation that runs through the whole GAN Zoo to be the most personally useful.  Jeff's presentation of the history of GAN development over time is really a great overview.  So feel free to skip ahead if some of the theory at the beginning that Mihaela is presenting is a little too abstract.


Observations:

1:  BigGANs (Brock et al.) talks about issue with batch processing on image net.  If not all classes are in batch, GAN can forget about it during training?

2:  Low res to high res techniques (Laplacian GAN, Progressive GANs)

3: StyleGAN introduces control of the latent space

4.  BiGANs, BigBiGANs seems same as another technique we recently discussed (CycleGAN?).  Dual generators working in opposite direction, discriminator looks at both, but they never communicate

5.  Pix2Pix (paired examples) to CycleGAN (non-paired examples)(cycle consistency)

6.  GANs for Video Synthesis  TGAN-v2, DVD-GAN, TriVD-GAN   spatial vs temporal discriminators

7. Program Synthesis - SPIRAL - DeepMind, learns to draw rather than just build image

8: Learning to see - artist using GANs

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