ai upscaling for everquest textures
 

this is an idea for the blue geniuses working on high def textures for titanium. i have no idea how to do any of this.

it looks like some modders have been using an ai machine learning algorithm to improve pre-rendered backgrounds for ps1 games like ff7 etc.

http://forums.qhimm.com/index.php?topic=18549.0

https://www.pcgamer.com/modders-are-...menal-results/

i thought maybe this could be applied to batch upscale textures from titanium to get quicker results than redoing each by hand in photoshop.

here is the program i think they use:
https://topazlabs.com/ai-gigapixel/

this is a cool idea so I am bumping

This is really interesting. It would be cool to hear about a machine learning algorithm that could upscale the polygon count for old school 3D models in combination with the textures that were designed to be applied to them .... I'm betting the result would definitely be better than Luclin models.

Yep, this is more or less the same concept behind EQClassicHD - originally it was just going to be a mostly automated upscaling project using waifu2x, which uses algorithmic methods self-described as "single-image super-resolution for anime-style art using deep convolutional neural networks". For our project, we found that a combination of algorithmic enhancement with some human finessing produced the best results, but as AI upscaling technology and computer technology in general continues to advance I'm sure superior results will be achievable.

"Machine learning" might not be proper terminology for this (although technically it is a form of machine learning or "deep learning"), since machine learning is usually characterized by a self-sustained iterative process like that of DeepMind, which essentially teaches and improves upon itself with insane results like achieving the equivalent of several hundred years of human progress in the span of a week. This is how AlphaStar recently went undefeated in a professional StarCraft II league, and how AlphaGo became the world's reigning Go champion, and so forth. Essentially what they do is teach an AI neural network the fundamentals of the game by letting it observe videos of real people playing, and then have the AI play against itself in a series of tournaments to produce a dominant AI "agent", which they then pit against another agent that observed different videos and played against itself in an entirely separate series of tournaments, etc. Pretty fascinating stuff, and humanity is doomed I'm sure.

Algorithmic upscaling kinda seems more like machine learning in microcosm, and a lot of the same concepts apply - it's inferring the increased detail during the upscale process, and I suppose it is iterative in that newer and better algorithms are being produced.

"trained on millions of images using deep learning technology" - a.i. gigapixel

pull whatever semantics you want from that. i just thought it looked cool and might be applicable to elfquest