0xArt
I am a french Deep Learning practitioner and PhD student. I make computational art with deep learnin
Calling the Thousand Torments
Pariah
Decadence
I printed out most of the images on photo paper, 20x30cm and two on canvas. They look DAMN GOOD.
Contact me if you want some printed out ;) ! Could be a nice gift for mother's day or father's day, if you told your parents that "uh, I got you something but the delivery is getting late" :D
Bones
A glitched world.
Funnily, this has been made by style transferring a glitched TV onto a picture instead of glitching that picture directly.
Castle in the Sky
I promise this page isn't turning into an Egyptian-themed bad museum photos gone cool thanks to neural nets.
It's just that I don't have any scheduled posts anymore and I'm late on pretty much everything.
I hope everyone is going well.
Despite being beatiful and super cool, generating pictures with style transfer is not very practical. It takes about 20 mins for a high definition image to be generated via the seminal work of Leon Gatys [1]. This doesn't leave a lot of room for iterative creative process and refinement. So obviously researchers tried to make that faster.
Johnson [2] proposed a very fast (almost real time!) way to transfer the style of one image to others, but it's far from perfect. It takes about eight hours to train a neural network to transfer ONE style. If you don't want to transfer the same style over and over again, that doesn't really help.
So I've been looking for ways to go faster with high quality pictures.
Yesterday I've been working on an implementation of AdaIN [3] in the hope that it works well enough. Below are some results while the net is training.
[1]https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gatys_Image_Style_Transfer_CVPR_2016_paper.pdf
[2] https://arxiv.org/abs/1603.08155
[3] https://arxiv.org/abs/1703.06868
Louvre
The cool thing with neural nets is that you can repurpose 10yo low res, blurry and crappy pictures and make them look good! Have you already repurposed an old pic for something that turned out good?
Sunset by the River.
Share your best sunset places or photos in the comments!
Green Hill Ireland Zone, Act 2, Dunluce Castle
In The Dark
Let's repopulate random forests with binary trees
New York City from the top
Notre-Dame in flames.
First picture generated in May 2020. Second one in April 2019. It seems my improvements and time spent on the algorithm and implementation were not for nothing.
London Parliament
That smile... That goddam smile
Ireland
Synthwave Toronto
They won't listen if you say it. Try screaming it instead.
Sunset Clouds Castle
8bits Toronto
The face of love
Food is the only true love. Food lets you eat it for your survival.
There once was a castle filled with tentacles
Hi everybody!
I play with neural networks & glitches to make art. Share this page if you like my work!
Most of what I create is made with my own implementations of Deep Dream, Neural Artistic Style or feature visualization algorithms open sourced in Torchélie. Glitch Art is made with GlitchLab.
Niagara from the side
Yosemite Park
Kaleidoscopic castle