Funny... we did something very similar for Jappeloup here at Hybride. We had 404 crowds shots to do in 5 or 6 different locations with different clothing styles. Some agents where to be seen very close to camera so we created high resolution geometry for the agents with ICE logic to mix various textures, clothing items, hair styles, etc. and we pre-baked a ton of cloth and hair simulations. We lined them all up on the timeline like you did and then artists put probabilities for each cycle appearing and it would randomly chose depending on the probabilities. We had about 50 animations cycles pre-baked for each man and woman agents. We started developing our deep compositing pipeline for this show since we though Arnold wouldn't be handle to handle all that high res geometry (some crowds where over 80 000 high res agents) but Arnold chewed everything up so we only finished our deep compositing pipeline a few months later for use on White House Down. We used the actual Ubisoft mo-cap studio to do all our mocap. We also created a 2D cards agent system as well with a few tricks to allow us to actually relight the footage in the cards. Those also gave very good results, but sometimes, having full 3D agents made it easier to integrate. It depended on the situation really. And we reused the same techniques for the Opera House in Smurf 2 as well and a very similar approach for our White House Down crowds. We don't have any making-of yet (we've been crazy busy for the past 2 years), but once we do, I'll gladly share. It's funny how various studios use similar techniques with the same softwares.
-Mathieu -----Original Message----- From: "Alan Fregtman" <alan.fregt...@gmail.com> To: "XSI Mailing List" <softimage@listproc.autodesk.com> Date: 09/13/13 13:01 Subject: ICE Crowds in "Now You See Me" (making-of/breakdown video) Rodeo FX has just put up a short reel of the crowds we did for "Now You See Me" to fill the MGM Grand stage with the help of ICE and Arnold... https://vimeo.com/74393635 It's not done with CrowdFX as SI|2013 was in beta while this was being made and they didn't need to be too intelligent, so we went with a bunch of nice cycle instancing tools and stationary particle instances. There were various variations of animation clips of various different variations of people. Animation was mocap captured with iPiSoft's playstation-eye-based mocap software, then cleaned up in MotionBuilder, brought back into Softimage (thanks to the MotionBuilder template rig) and caches exported out. The cycles were in one long timeline of one clip after another and we stored start & end frame numbers along with an array of ICE strings (of the cycle names.) We might have "clappingA", "clappingB", "clappingC" with different frame ranges and then we had a neat ICE compound where you could give it in a substring (eg. "clapping") and it would find all variations for that name and randomly assign those frame ranges and cycle. If I recall correctly the general behaviours were: standing idle looking around, clapping normally, clapping hyperenthusiastically with bonus fistpumping, and grabbing money bills from the air. There were three or so variations of each. Furthermore, the crowd on the floor near the stage is CG, but the one in the stadium seats is actually 2D cards of footage of real people -- Rodeo employees, in fact -- doing various motions, instanced in Nuke with some scripted magic. (I was not involved with the 2D crowd so that's as much as I know.) The 3D crowd models are Rodeo folks too, by the way. I'm among them, as are most of my coworkers. We used some software with the Microsoft Kinect to get some general 3D scan meshes of us as a reference for volume/form, but they were modeled by hand as the scan wasn't quite perfect as-is. It was super helpful to have the scans though! Its pretty amazing how often you can tell people apart from their silhouette/stance alone. I co-developed the ICE side of it together with Jonathan Laborde (who is in the list and probably reading this.) Hats off to my other fellow coworkers who modeled, textured, lit and comped everything so well. :) Teamwork! Cheers, -- Alan
