September 11, 2008

Radiohead’s city street

Available to down­load from Google is a small part of the data set from Radiohead’s recent House of Cards music video, shot entirely with­out cam­eras — just LIDAR and 3D cap­ture. Part of the idea is that fans will be able to remix the “footage” or mash it up in some way.

This is tremen­dously cool and very thought­ful of Radio­head to copy­left this part of their work.

Unfor­tu­nately Google or Radio­head has only pro­vided a one minute sample of the ani­ma­tion data, which isn’t much to work with. Of the 14,000 data down­loads and count­ing, I think this prob­a­bly explains why no one has come up with any­thing too strik­ing on the YouTube Remix Group.

Still, I grabbed the data and got to work. First step: better tools. Per­son­ally I feel the Pro­cess­ing code is just too slow and painful to work with. OpenGL really needs to be run through native code.

So the first app I wrote is a Obj-C based first-​person browser of the city street land­scape (which includes 1.3 mil­lion ver­tices). It’s intel Mac-​only and com­pletely untested, but if you’re daring it might be fun to play with. It takes a couple sec­onds to load all the points, but then runs sur­pris­ingly fast. I get 30fps on my first-​generation Mac­Book Pro.

hoc-teaser.png

You can down­load it here (21.6MB):
http://​files.​car​son​baker.​org/​h​o​c​/​H​O​C​-​C​i​t​y​S​c​e​n​e​V​i​e​w​e​r.zip

Or watch a quick demon­stra­tion on YouTube.

Two more apps are on their way. One just dis­plays the other land­scape; the other is a facial ani­ma­tion viewer of Thom. These projects are lay­ered on top of a custom OpenGL threaded graph­ics frame­work I built four years ago — just some­thing that comes in handy from time to time. In a way it has kind of evolved into its own mini-​version of Pro­cess­ing, with­out the per­for­mance penalties.

So, what about source code? Well, if Radio­head releases the entire data set then I’ll post sources for every­thing, includ­ing my graph­ics frame­work and the client code. I don’t know how hard it is to pre­pare the data, but I don’t think that’s too unrea­son­able of an idea. I think it’s win-​win — more people could put together better mon­tages with a more capa­ble tool.

If Google decides to release more of the data, one thing I sug­gested is not dis­trib­ut­ing the vertex points as CSV data. It makes it much bigger to down­load and slower to parse. Why CSV anyway? Hope­fully people aren’t open­ing this stuff up in Excel. Save it as binary! We can handle it. Here’s some code to do it:

http://​files.​car​son​baker.​org/​h​o​c​/​c​s​v​_​t​o​_​b​i​n​a​r​y​.​c.txt

Also — for anyone that might know:

It’s not clear to me how the Geo­met­ric Infor­mat­ics cap­ture hard­ware works. What’s the tech­nol­ogy behind it? How much does their GeoVideo system cost? I’m having a hard time find­ing this out.