The Power of Cartoonization: A Breakthrough in Image Processing Technology

As AI continues to advance at an unprecedented rate, researchers from the popular S2 Paints research team has just released a groundbreaking paper that can cartoonize your pictures and videos in the style of a specific cartoon. This innovative technology is called image cartoonization, and it uses a novel framework to learn from the cartoon style behavior.

What's Cool About this New Method?

One of the most impressive aspects of this new method is that it's separately learning each extracted representation. This means that the framework can be made controllable and adjustable to better match the artist's style in different use cases. To achieve this, images are first decomposed into three cartoon representations, which serve as a guide for the network's optimization to generate the cartoonized results.

Image Decomposition and Representation

The decomposition process involves breaking down images into three distinct components: surface structure, texture representation, and another structure representation. The first component is used by a discriminator (ds) to distinguish between the surface representation of the outputs and the original cartoons. This helps create a more realistic cartoonish style.

"WEBVTTKind: captionsLanguage: enthis ai can cartoonize any picture or video you feed it in the cartoon style you want let's see how it does that and some amazing examples you can even try it yourself on the website they created as i did for myself this is what's ai and i share artificial intelligence news every week if you are new to the channel and want to stay up to date please consider subscribing to not miss any further news researchers from the popular s2 paints research team has just released a new paper able to cartoonize your pictures and videos in the style of a specific cartoon this is called image cartoonization it uses again framework in order to learn from the cartoon style behavior what's cool about this new method is that it's separately learning each extracted representations making their framework controllable and adjustable to better match the artist's style in different use cases to do that images are first decomposed into three cartoon representations which will guide the network's optimization to generate the cartoonized results it separately takes the surface structure and texture representations in order to produce equivalently realistic cartoonish results for all these components as said it uses a gun framework but with one generator and two discriminators where the first discriminator ds aims to distinguish between the surface representation of the outputs and the original cartoons the second one dt is used for the same goal replacing the surface by the texture of the images then a pre-trained vgg network extracts high-level features and imposes spatial constraint between the structure representations of input photos and outputs if we dive a bit deeper into the three principle segments we can see that the surface representations role is to imitate the cartoon painting style made possible by the use of the discriminator ds as discussed previously then the second principal component they use is the structure this is where the vgg is used it extracts the high level features to enforce spatial constraint between the results and extracted structure representations they use a standard super pixel algorithm to find the segment maps using these maps to color each segmented region of the image with an average of the pixel value finally the texture representation is used in the form of a random color shift to extract single channel texture representation from color images in short it returns high frequency texture information and decreases the influence of color and luminance to make the results look more cartoonish it is also super customizable you can decide the sharpness of the texture just by fine-tuning one parameter which will easily make it look the way you want just take a minute to look at the awesome results they got you can even do it on a video here they applied it on an avengers scene look how sharp and clear the results are fellow done me wrong got a lousy taste it meant kid he's not so bad he has a temper deep down he's all fluff it's nice what what what is you and romanov no we haven't that wasn't it's okay nobody's breaking any bylaws it's just she's not the most open person in the world but with you she seems very relaxed no natasha she's she likes to flirt i've seen her flirt up close this ain't that please start by all means here we go okay never one to shrink from an honest challenge it's physics right so if i lift it i i then rule ashgard yes of course i will be reinstituting primonocta be right back are you even pulling are you on my team just represent paul all right let's go here's the result i personally got on myself just to see that this really works on any picture it takes i invite you to try it yourself on a picture or a video using the first link in the description of the video the code is also available in their github if you'd like to train it on a different cartoon style of course this was just a simple overview of this new paper i strongly recommend to read the paper linked in the description for more information please leave a like if you went this far in the video and since there are over 90 percent of you guys watching that are not subscribed yet consider subscribing to the channel to not miss any further news clearly explained if you want to support the channel i now have a patreon linked in the description where you can do that thank you for watching youthis ai can cartoonize any picture or video you feed it in the cartoon style you want let's see how it does that and some amazing examples you can even try it yourself on the website they created as i did for myself this is what's ai and i share artificial intelligence news every week if you are new to the channel and want to stay up to date please consider subscribing to not miss any further news researchers from the popular s2 paints research team has just released a new paper able to cartoonize your pictures and videos in the style of a specific cartoon this is called image cartoonization it uses again framework in order to learn from the cartoon style behavior what's cool about this new method is that it's separately learning each extracted representations making their framework controllable and adjustable to better match the artist's style in different use cases to do that images are first decomposed into three cartoon representations which will guide the network's optimization to generate the cartoonized results it separately takes the surface structure and texture representations in order to produce equivalently realistic cartoonish results for all these components as said it uses a gun framework but with one generator and two discriminators where the first discriminator ds aims to distinguish between the surface representation of the outputs and the original cartoons the second one dt is used for the same goal replacing the surface by the texture of the images then a pre-trained vgg network extracts high-level features and imposes spatial constraint between the structure representations of input photos and outputs if we dive a bit deeper into the three principle segments we can see that the surface representations role is to imitate the cartoon painting style made possible by the use of the discriminator ds as discussed previously then the second principal component they use is the structure this is where the vgg is used it extracts the high level features to enforce spatial constraint between the results and extracted structure representations they use a standard super pixel algorithm to find the segment maps using these maps to color each segmented region of the image with an average of the pixel value finally the texture representation is used in the form of a random color shift to extract single channel texture representation from color images in short it returns high frequency texture information and decreases the influence of color and luminance to make the results look more cartoonish it is also super customizable you can decide the sharpness of the texture just by fine-tuning one parameter which will easily make it look the way you want just take a minute to look at the awesome results they got you can even do it on a video here they applied it on an avengers scene look how sharp and clear the results are fellow done me wrong got a lousy taste it meant kid he's not so bad he has a temper deep down he's all fluff it's nice what what what is you and romanov no we haven't that wasn't it's okay nobody's breaking any bylaws it's just she's not the most open person in the world but with you she seems very relaxed no natasha she's she likes to flirt i've seen her flirt up close this ain't that please start by all means here we go okay never one to shrink from an honest challenge it's physics right so if i lift it i i then rule ashgard yes of course i will be reinstituting primonocta be right back are you even pulling are you on my team just represent paul all right let's go here's the result i personally got on myself just to see that this really works on any picture it takes i invite you to try it yourself on a picture or a video using the first link in the description of the video the code is also available in their github if you'd like to train it on a different cartoon style of course this was just a simple overview of this new paper i strongly recommend to read the paper linked in the description for more information please leave a like if you went this far in the video and since there are over 90 percent of you guys watching that are not subscribed yet consider subscribing to the channel to not miss any further news clearly explained if you want to support the channel i now have a patreon linked in the description where you can do that thank you for watching you\n"