Using all the information you get in a DAE file i modified my away3d a bit to auto place the camera, set the focus, adjust the fov, etc. The material is applied and the scene is played out at the specified frame rate. Each frame a screen capture is done (we got the cropping data from a photoshop plugin to reduce the size of the capture) and is stored in a vector. The video is played back on a timeline and calls are made when a model is needed to be played back with the video. A bitmap is placed behind the video and for each frame the bitmapData is updated from the vector created by the renderer. The reason i called it compositing is because using just the DAE file i lined everything up perfectly with the 3d scene / video it was tracked to. Ran into a lot of issues along the way like away3d is a yUp world like 75% of the 3D apps out there. 3DSMax is a zUp world which the solve for this is to switch the rotation values for z and y. This works individually but with multiple you get into rotation order messiness. To fix that you have to adjust the separate the quaternion into 3 different ones then adjust the order in which they are multiplied. This converts your rotation order to yzx instead of default zyx. This sort of brings you to the reason why DAE files are slow. They use rotation values in degrees instead of quaternions or matrices (also vulnerable to gimbal lock). Im not too familiar with MD2 but what does it use for its rotation info? Anyway back to compositing. We also used LUTs for curves and saturation to colorize each scene separately to compensate for lighting changes. We can't get every shot perfect due to the lighting coming from one side of the face or another but we feel like we did a fairly good job with it. So all of the 44 scenes were controlled from an XML file which made my life a bunch easier. Sorry i think i started rambling a bit...
On May 3, 8:28 am, Páll Zoltán <[email protected]> wrote: > niiiiiiiiice-nice job extralush, that's very cool! :) > the 3d face is composited on the frontend? it seems like its done on > backend, but that begs the question "why is this thread called awat3d video > compositing" > i'd really appreciate a short walk-through > > thanks, > Z > > > > > > > > On Tue, May 3, 2011 at 2:00 PM, Fabrice3D <[email protected]> wrote: > > 45%... 46%... > > the compositing is kinda slow. I want to see the result so I let it run in > > background :) > > but the baby is already like 15 minutes busy doing "things" > > > I think you should add something, like "come take a free coffee at the > > hotel bar" video if you do not want to loose your visitors > > I also think you could offer to register the email, and add a "send me > > email when ready" > > > the head displayed as example in first edit screen is bit high on Y axis, > > and do not allow rotations, this would help to map a bit better. > > > oh now 70% almost... :) > > > Fabrice > > > On May 3, 2011, at 12:31 AM, extralush wrote: > > > > Hey guys. We launched a site today that has 3D face upload action > > > using Away3D. 12 different videos and roughly 44 different scenes > > > tracked and composited to video. We exported or files from 3DSMax > > > using openCollada. Ran into plenty of weird things to get this working > > > in an automated way and will eventually make a blog post or seven > > > about it. Check out the link and let me know what you think. Upload > > > your face to all 4 characters to see them all. > > > >http://www.tripyourface.com/ > > > > If you have any questions or comments on the 3d / compositing portion > > > please feel free to IM (willbreel) or Skype (willbreel) me.
