motion buffer strangeness (or: frame buffers again :))

[riviera]

Hi,

I'm trying to use the motion buffer output of mental ray (3.4.6.18) and having some strange results.

It seems that the motion buffer doesn't contain motion information for pixels at the objects' contours. Compared to main image and z-buffers, the motion buffer always seem to have about 1 pixel 'shrunken' than any others.

Anyone any ideas?

imre

[bart]

Do you mean the motion vector framebuffer?

Are you using non-interpolated z depth framebuffer, ie "-z" in mi scene description to use the nearest sample for the pixel?

There was a patch to make the motion vector sample correlate to the nearest z depth sample in that case later in 3.4. Don't know if it made it into 3.4.6.1.

[riviera]

I'm using 3.4.6.18 (the forum engine translated 8 and ) to a smiley 8))

Yes, I mean the motion vector framebuffer.

You suggest the motion vector framebuffer contents would be different if I used different interpolation for the z-buffer?

I haven't tried changing interpolation on the z-buffer yet (-z/+z); what I tried though is -m/+m, but to no avail: it gave the same results.

When I extracted the motion vectors using a regular rgb custom buffer (or using the regular main image buffer, for that matter) I got the proper results as expected, of course.

However I rather use the actual motion buffer to get the motion data, instead of doing workarounds and such.