w00t

OptiX for DirectX Programmers: Raytracing Eve Online's Portraits

Sat, 12 May 2012 14:08:20 +0000

I'm off to the GPU Technology Conference 2012 in San Jose to talk about our experiences with integrating Nvidia's OptiX package into our existing DirectX engine. Here's a teaser, I will upload the slides after the talk :)

WoD - Animatic Trailer

Sat, 12 May 2012 00:25:39 +0000

Huh, I just discovered that around Fanfest we also uploaded the WoD animatic trailer from some time ago. This has been online for a while but it was a shaky-cam version with some bald dude in the front. Nice that we now get to enjoy it in proper quality :)

Eve in Space

Thu, 10 May 2012 20:54:15 +0000

Awwww yeeeeeah.

Mobile Flickr

Tue, 8 May 2012 21:19:55 +0000

Next week I'm off to a conference, so I just wanted to spam again the link to the flickr stream of my mobile phone, just in case I do anything besides sitting in a dark conference hall, geeking out \o/

MLP - Earthquakes visualized

Tue, 17 Apr 2012 23:07:20 +0000

This is pretty cool!

15 Years of Icelandic Earthquakes and Eruptions in 4 Minutes from hjalli on Vimeo.

It seems like it got worse again after I got here.. maybe I should take road speedbumps a little slower.

Chill Beats I-V

Wed, 11 Apr 2012 18:12:29 +0000

Great for programming \o/ Or staring at Iceland.

Harmonic 33 - ExtraOrdinary People

The Peace Orchestra - The Man, Part One

Boards of Canada - Dayvan Cowboy

Cornershop - State Troopers, Part I

State Troopers, Part 1 by Cornershop on Grooveshark

Pelpp & A.Vanvranken - Waste

Waste by Pelpp & A. Vanvranken on Grooveshark

Stochastic Rasterization using Time Continuous Triangles

Sat, 7 Apr 2012 15:17:55 +0000

Easter Break + really shitty weather = time to write some code :D

Stochastic Rasterization using Time Continuous Triangles, and the paper that it extends, Triangle scan conversion using 2D homogeneous coordinates, have been on my coffee table since forever. I finally got around to doing an implementation of its core to check out stochastic motion blur.

This test clip uses a deliberately low 15-FPS to compare point sampled and therefor jittery video (left) with 16xAA stochastic blur (right).

This is a fast moving triangle that also changes color as it goes back and forth. The clip shows two effects that you cannot do with a full screen postprocess that's popular these days. First, the movement changes the visibility dramatically; a simple blur would just smear out the triangle or the checker-board, instead of correctly revealing, and blurring in, the squares behind the triangle that the motion reveals. Second, blurring would keep the color fixed, instead of mixing the time-varying pixel shader output.

To be fair, that last point means you can't technically talk about "MSAA" anymore; it's actually SSAA, full shader re-evaluations required.

Here's a closer look; the top row shows the position of the triangle in frame N and frame N+1. The second row shows the output of a typical accumulation buffer approach, with uniform samples. The banding is obvious, and the stochastic approach trades this for noise. At such high speeds it's still not pretty at 4xMSAA, but 16xMSAA starts to look decent.

Testing rotation:

Camera pan:

The next step would be to port this to Cuda/DirectCompute; and add a stream-processor compatible tile binning step so performance is acceptable with real world triangle meshes.

The paper also talks about how to do shadowing in a setup like this. The problem is that the randomly chosen time samples in the shadowmap t_S don't match the time t_Q of the query point. If the triangle is moving towards the light and it happens that t_Q < t_S, then the point in the shadow map will already be closer to the light than the point you're shading, and you will get incorrect self shadowing. Likewise if the triangle is moving away and t_Q > t_S. Even when it's guaranteed that t_Q and t_S are in the same stratified interval, as the authors suggest, chances are fifty-fifty to get an incorrect "in shadow" answer -- so even with PCF this will converge to the wrong 50% grey.

One possible solution could be to ignore t_Q and instead use t_S to recompute the depth of the query point, using a raycast on the moving triangle. Doesn't sound cheap, but it might work. That, or I'm missing something :)

Icelandic Winter

Tue, 3 Apr 2012 23:50:46 +0000

What can you do in one hundred hours? Learn Icelandic? Learn to play the piano? Learn how to draw? Practice pick up lines in Islenski Barinn?

I should probably stop paying for this website, I'm out of stuff to say :\

Eve / Dust Trailer

Sat, 24 Mar 2012 19:34:44 +0000

A new fanfest, a new cinematic trailer. Probably our best trailer so far, crank it way up and enjoy :)

Edit: found a link to the bombing demo :)

Eve Online Fanfest 2012

Thu, 22 Mar 2012 20:49:38 +0000

Today was the first day of Fanfest, and it was splendid. Here's some shots from Harpa, the new venue.

The next shot is from a Rifter that one of the players made out of Lego. The model was submitted to CUUSOO; if it reaches 10000 votes, someone from Lego looks at it to determine the possible commercialization of the idea. It would be pretty cool for sure :)

Another view, in the background is the Eve Store selling memorabilia and slick apparel.

Today was basically "Dust 514" day, and the keynote was grand. We showed a live bombardment from Eve into Dust, no cheating and no tricks, and it was brilliant. Never done before, up until today that is. Go CCP :)