Interesting graphics and animation research

This page is a list of interesting,  maybe not-quite-ready-real time graphics and animation tech that I'm keeping an eye on. Should be growing over time.

Rendering

Anybody in games should be interested in the fate of Ptex, the no-uvs, variable resolution texturing system from Disney and used in, for example, Mari.  So far, it's not ready for runtime graphics. but this Nvidia slideset discusses a realtime implementation.  It claims to be only 15% slower than a UV mapped alternative - while saving about 18% of texture memory.  I'll believe it when I see it, but i want to believe. ( Update: I went over this with our residence graphics genius and he doesn't think it'll fly, even on XBone / PS4 class hardware)

A GPU Gem on doing hardware aging of materials using GPU particles. Can't wait to see this get out from behind the paywall!

Joint Importance Sampling: a paper from Disney on a better way to render path-traced media like fog or water.  Claims to increase render times 1000x for little change in quality! And it's still too slow for games :)

A new(ish) paper from Peter-Pike Sloan proposing a method for low-res, realtime friendly pseudo gi lighting.  The new hardware generation is not going to make realtime GI happen with the current techniques, so new ways of looking at GI are important.

Also from PPS and Bungie, a review of the hardware-based AO bake system used on Destiny.  Note that in this example "AO" is "ambient obscurance" rather than "ambient occlusion" - obscurance accounts for distance to light sources, where as tradtional occlusion is dimensionless.

A neat paper from Sebastian Lagarde on rendering wet surfaces in a physically based environment.  Seb's blog is a top-notch reference for anybody doing Physically based rendering (which means most of us by about 2015)

Animation

Motion fields  "Motion graphs" - basically efficiently seaming together  a desired set of Mocap data  - has been a big deal in academia for a long time, but it's never caught on in games due to the Prince of Persia problem: it's too realistic and thus not responsive enough. This work from the graphics group at UW is an interesting slant on making this more reactive.  Unfortunately @ZoranPopovic has largely moved on to education from animation, but his group's page at UW has a lot of interesting work from the last 5-6 years.

I'm a huge fan of Kevin Wampler's work on procededural animation. This paper from 2009 describes a (slow, offline, but cool) process for generating decent gaits for arbitrarily shaped creatures. Check out the videos (note: big download, requires divx).  This one from the Netherlands is a similar optimization based approach to bipeds only - the big wrinkle being that everything in this one runs on simulated muscle networks rather than pure joint motors. Fun stuff!

This is an interesting paper from Karen Liu's group at Georgia Tech, featuring soft body simulation done with a combination of finite element modeling and a muscle fiber system. Soft body animation remains a pretty black art, it would be great if it catches on in realtime-friendly ways. Not holding my breath, alas.

This talk from the developer of Overgrowh is an interesting look at hybrid physical-kinematic animations, a sort of  'Euphoria Lite'.  Some related references from AIGameDev.  Also related is Rick Lico's talk from GDC 2014 about the Destiny animation system (GDC vault paywall).

Procedural worlds

I'm very interested in procedural worlds for obvious reasons.  It's an area that's got enornous potential but has been moving very slowly in the game side of things.

The early chapters of this dissertation provides a decent overview of lots of the terminology and existing work in the field

There's CityEngine,  which looks good if you're looking at city size data sets (and it supports python!)  There are a lot of nice procedural terrain packages but they all look very fractal-y to me  nowadays; it was impressive in Wrath of Khan but hasn't moved on much since. The most interesting terrain paper I've seen in a while is this one, which uses rivers as the structural basis for terrain generation..

A collection of papers from Purdue on procedural land use - thing like zoning and road layout. Some of this looks pretty handy.   If it gets scary, just remember 'markov chain'  is math-speak for 'state machine', more or less.  The other fun aspect is their efforts to  'inverse model' the stuff they study - to turn real data into the data their model uses.

Meshes, models, and surfaces

This paper from Columbia promises an optimal way of establishing surface flow across arbitrary surfaces. This is interesting for a couple of reasons - it might be an interesting improvement to mesh-based navigation, and it might also lead to alternate methods of automatic UV generation. 

A decent roundup of mesh parameterization techniques