Look at this video i made about a month ago that turns real life scenes into anime, like style. Well today, when you finish watching this, video youll be impressed even more as we have ourselves a new ai to play with previously, i experimented with cartoon game, which is an older, deep learning algorithm. Well, not really that old and trout on different anime scenes that were inspired upon real life sceneries and the result was not that convincing. But the funny thing
Is soon after i published this video, a new research paper popped up, and, incidentally, caught my eyes, its called white box, cartoonization and that’s. What were talking about today? The main problem that cartoon gan has is that it contains a lot of artifacts and generates in a very low resolution, but white ball cartoonization came into this competition and destroyed everyone else. It can generate naturally in a very high resolution and well, it still produces some artifacts, but its much less than any other deep learning architecture out there, and the question here is: how is this achieved? Well, the sure
answer is the ai considers more aspects in terms of the input video the more control we have when training the ai the better it will be basically if you want to know the long answer it is well documented in your blog for ais is transforming real life footage into the lets go this anime style footage the goal or the product is to simply simplify the features of objects produce less complex shapes and have more precise edges when perceived from a video they are the process the product of transformation is usually cutting down.
some fine details like human faces which are far too detailed or the small prints on the boxes or cups which is removed well more like simplified just look at this underground entrance it looks pretty good right also the reflections on the car is pretty amazingly retained in the ai did not create any artifacts surprisingly as some of the reflections are pretty distorted for this shot.
outline of the drone is crisp and the parts where its supposed to be blur is blurred which is pretty well done here so at this point i think we pretty much already proven that white box cartoonization is well much better than cartoon game but the real question here now is that how similar is the output from white box cartoonization compared to the animes that uses real-life sceneries like maybe is there a possibility where this can replace a part of the anime.
production well probably no for now but in the future there are probably chances when this technique is more developed while its mainly because now drawn is still way better looking than ai processed in the meantime i also got curious about how cartoonization works on different other videos like this gaming montage i made for cs go a long time ago which seems like a pretty cool concept and it actually worked pretty well well my curious mind also told me to test it on some animes.
so i chose violet evergarden because of their amazing art style and textures well what we can learn from this is the ai is programmed to take in any video and simplify it so it doesnt matter if you input animations or anything more simple the ai will always give you the simpler result and i also took this video and attempt to cartoonize it and shockingly the cartoonization is really really well i was expecting some more artifacts but it was actually pretty crisp it looks like one of those early age.
animation to an extent which is pretty pretty pretty cool well that’s it for today now um thank you guys for watching and here are some bonus cartoon nice clips if you havent follow me on twitter please do and ill see you all next time.
Leave a Reply