Re: [Openexr-devel] "standard" for storing a depth map?

[Christian Bloch]

I agree with Christopher. Normalizing is a bad idea, because it may not be consistent throughout the entire sequence. When you have a distant object travel through frame, or simply pan from an empty sky area to a cityscape the Z buffer will pump back and forth.

It's also unnecessary with 32-bit float.

We're saving the straight distance values. For the sake of a good default I configured our own pipeline with a factor of 0.1, just so a distance of 10 meters becomes a value of 1. So far that seems to be working great for most scenes.

I did setup Normalizing in an alternative pipeline (where Z is rendered separately for better AA control), but that requires the artists to visually pick the near and far values using OpenGL fog, and those values are kept consistent across the sequence.

.Christian Bloch

On Mar 14, 2012, at 2:41 PM, Christopher Horvath wrote:

If you're using 32-bit float for depth, why do you need to normalize at all? Doesn't that defeat the purpose?

Oh good idea. I'll be normalizing my values, but everything is float.