Ian Goodfellow: David has released a new paper from an old collaboration. Glad to see it out!
3 replies, 498 likes
roadrunner01: Creating High Resolution Images with a Latent Adversarial Generator
github(not yet updated): https://github.com/google-research/lag https://t.co/KwnIGn67pK
3 replies, 121 likes
Peyman Milanfar: 1/ 3 Instead of one high res output, we generate a family where the low res input only guides a "subspace" of images the network can produce. Latent Adversarial Generator (LAG) learns in latent space of the adversary w/ a perceptual loss & w/o a pixel loss
2 replies, 111 likes
Peyman Milanfar: Update: Code is now available for our Latent Adversarial Generator work.
0 replies, 35 likes
Colin Raffel: This paper, led by the incredible @D_Berthelot_ML, has super-impressive super-resolution results that have been kicking around for a while (I included them in this talk from last year: https://colinraffel.com/talks/nyu2018why.pdf)
1 replies, 15 likes
David Berthelot: New paper with @docmilanfar and @goodfellow_ian
0 replies, 11 likes
arxiv: Creating High Resolution Images with a Latent Adversarial Generator. http://arxiv.org/abs/2003.02365 https://t.co/iNLLq2Enha
0 replies, 3 likes
Maurice Lee: Woah. This is pretty neat and seems relevant for the "super-resolution microscopy + machine learning" people!
Instead of generating only one unique image as the output, they have a possible set of related images.
Figure 7 seems to be the stuff of nightmares. https://t.co/lUtOmFUI6Y
0 replies, 2 likes
arXiv CS-CV: Creating High Resolution Images with a Latent Adversarial Generator http://arxiv.org/abs/2003.02365
0 replies, 1 likes
Found on May 21 2020 at https://arxiv.org/pdf/2003.02365.pdf