It looks like Neural Style Transfer to me, they've probably taken a selfie then used a neural network to repaint it in the style of another image (probably a surrealist painting) either programatically (it's not as hard as it sounds) or through an app.
Whoa. Is that a “for dummies” version of Tensorflow that uses a programming language suitable for non-programmers?
Sadly, no. Neural Style transfer is the name for using a neural network to take the content of one image (in this case the selfie) and make it look like it was created in the style of another image (the surrealist painting). What I linked is a tutorial how to do that in Tensorflow, which is an AI & Machine Learning library for Python & C++.
The actual algorithms and neural networks designed to do it are incredibly complicated, and no-one who doesn't have a post-graduate qualification in computer science is expected to understand them but thanks to the good people at tensorflow and the kind person who wrote the tutorial you don't need to understand them to do some really cool stuff with image style transfer. You just need to install tensorflow, copy the code from the tutorial, and you too can see what it would be like if instead of a starry night, Van Gough was inspired to paint in the same style a pig defecating on its own balls.
It's really really fun and if you know how to install and work with libraries in Python and are looking for a coding project I can strongly recommend getting a neural style transfer network running on your own machine.
If you're not a programmer and not interested in learning programming then don't worry about it, but basically, take the content of any image and make it look like it's in the style of any other image.
So you can take a photo (like this guy did) and make it look like it's a surrealist painting, or any other style of painting, or a cartoon, or from a video game. Or take a cartoon or videogame screenshot and make them look like a photo or painting. Once you've got you're neural network up and running all you need to do is choose the image you want it to get the style from, the image you want it to get the content from, and maybe play around with some parameters to get a result you're happy with.
That's the theory anyway, results vary based on the two images, your processing power, and how suited to those images the neural net you've chosen to use is, but they can often look really good.
I started a thread - https://hexbear.net/post/159260
—————————————————————
I found an online site to play around on. I merged an eye and part of a koi. Playing around in raw code is no fun for me but I'm tempted now. Still - I have to ask - do you know of a simplistic "mashup" like Photoshop filter on steroids that would - for example - let me "blend" the pattern of the koi into the eye?
My end goal is to just fool around with a source photo into the I'll call it a source "pattern/color". I'm not trying to get Van Gogh to come out.
It looks like Neural Style Transfer to me, they've probably taken a selfie then used a neural network to repaint it in the style of another image (probably a surrealist painting) either programatically (it's not as hard as it sounds) or through an app.
Whoa. Is that a "for dummies" version of Tensorflow that uses a programming language suitable for non-programmers?
For me - learning even baby Python and baby C++ is a 10/10 level of difficulty.
Sadly, no. Neural Style transfer is the name for using a neural network to take the content of one image (in this case the selfie) and make it look like it was created in the style of another image (the surrealist painting). What I linked is a tutorial how to do that in Tensorflow, which is an AI & Machine Learning library for Python & C++.
The actual algorithms and neural networks designed to do it are incredibly complicated, and no-one who doesn't have a post-graduate qualification in computer science is expected to understand them but thanks to the good people at tensorflow and the kind person who wrote the tutorial you don't need to understand them to do some really cool stuff with image style transfer. You just need to install tensorflow, copy the code from the tutorial, and you too can see what it would be like if instead of a starry night, Van Gough was inspired to paint in the same style a pig defecating on its own balls.
It's really really fun and if you know how to install and work with libraries in Python and are looking for a coding project I can strongly recommend getting a neural style transfer network running on your own machine.
What could I possibly do? I'm not a programmer and it seems like a total nightmare.
If you're not a programmer and not interested in learning programming then don't worry about it, but basically, take the content of any image and make it look like it's in the style of any other image.
So you can take a photo (like this guy did) and make it look like it's a surrealist painting, or any other style of painting, or a cartoon, or from a video game. Or take a cartoon or videogame screenshot and make them look like a photo or painting. Once you've got you're neural network up and running all you need to do is choose the image you want it to get the style from, the image you want it to get the content from, and maybe play around with some parameters to get a result you're happy with.
That's the theory anyway, results vary based on the two images, your processing power, and how suited to those images the neural net you've chosen to use is, but they can often look really good.
Edit
I started a thread - https://hexbear.net/post/159260
—————————————————————
I found an online site to play around on. I merged an eye and part of a koi. Playing around in raw code is no fun for me but I'm tempted now. Still - I have to ask - do you know of a simplistic "mashup" like Photoshop filter on steroids that would - for example - let me "blend" the pattern of the koi into the eye?
My end goal is to just fool around with a source photo into the I'll call it a source "pattern/color". I'm not trying to get Van Gogh to come out.