Stereoscopic compositing

[Anonymous]

Hi all

I am trying to find advice/solutions for stereoscopic compositing of greenscreen elements in 3D space.
What happens with shadows for the greenscreen elements?
How do we line them up and cast them effectively on a stereoscopic filmed backplate?
Has anyone had any other nasty surprises with stereoscopic compositing?
Are there any resources/links for info?

Cheers
Craig

[amit Dang]

Hi. What application are you compositing in? I’ve developed a few different techniques for stereo compositing, but he’s a basic work flow:

If you have good shadows on the green screen just use a different keyer to pull those separate form your main matte. When the shadows came from multiple light sources and did not work, we had to roto or create CG roto-mation based shadows to comp in.

In the comp I always use a setup that creates a boolean or channel swap operation that creates an anaglyph image to check stereo as the comp progresses. You do this by desatuating each L and R eye image to a grey scale then taking the red channel from the L eye and replacing the red channel of the R eye.

Then on each comp layer ( backplate, greenscreen, CG layers Etc) I create a transform node on the Left Eye that is locked on the X axis and it is linked via expression to an inverse of the the X offset fed into another transform on the Right Eye. You can now slide the comp layer in visual Z depth space while viewing the anaglyph’s output.

Adjust that until the elements look right.

Basic rules of thumb:

Never break frame when a layer is in front of the parallax coming off the screen.

Things that recede into depth should never ever have more than 5 cm distance apart on a projected screen. The human eye goes into orthographic stereo after about 150 feet, it does not go wall eyed.

Always make sure elements like feet or cars or anything touching something else is on the same plane in depth. It may look right in 2D but 3D reveals all.

Also pay attention to the interocular distance and the scale of 3D objects. you must build to scale or scale your interocular to match. The only reason a miniature model looks real in effects for movies is the loss of 3D depth. In stereo, you can sense scale of objects. you must compensate for this.

Cheers,

Daniel Smith
VFX Supervisor

[jonmason]

Thanks for the reply, much appreciated.
We will be comping in Nuke or Shake. Getting things to stick to the ground is probably our main concern as there are going to be numerous elements to comp in any single shot.
Locking the shadows to the same point of ground for L & R images could be interesting (I predict a little warping here so that we can match the terrain of the bg)

I would be interested to know what kit you have been using/ whether it was suitable for the job?

Cheers
Craig

[amit Dang]

I’ve used both Nuke and Shake on Stereo jobs. I have also used Fusion. They are all very capable of doing a great job, but the work flow is pretty standard as I outlined above. Checking the comp in 3D is a must. Also being able to slide the elements in Z-space using the Transform nodes with the inverted expression is a must also.

-Dan

[Author]

Posts