-
RSMBvector and FG/BG motion blur depth sorting.
How do you guys deal with crossing object and RSMBvectors ? i always have horrible results at the border between two huge differences in motion intensity and/or direction (typically when two objects cross), as RSMB doesn’t know which one is in front or behind ?
For example, frischtluft Depth of field is able to rebuild some covered area of the image if a foreground part need to be blured by the depth of field (and thus should reveal an inexistant part of the bg). Then it do some depth compositing given the fact that some part of the foreground should be more blury than the BG or the opposite.
In my opinion, RSMB should do the same, because at the moment it doesn’t carry information about what is in front, and what is behind. It just have velocity information about pixels, but doens’t know how to depth sort its strokes of blur.
If you have tips, i would love to ear it.
-
3 Replies
-
Hi Baracuda,
For now you sometimes have to split your render into different passes. In some cases (eg the background does not move), you can simply key out zero motion (0.5, 0.5, 0) off the Velocity image to avoid edge artifacts.
We do discontinuity detection in RE:Map for UV mapping but it would not work as well for motion blur.
Pierre
-
-> For now you sometimes have to split your render into different passes
yep but it’s pretty complicated when you have a complexe geometry in movement, for example, a body walking, with crossing legs and arm passing in front of the hips. I can’t split my character into smaller part, like left and right legs.-> In some cases (eg the background does not move), you can simply key out zero motion (0.5, 0.5, 0) off the Velocity image to avoid edge artifacts.
when i do this, it looks like if my character is motion blured, but then cutted out through it’s non blured alpha. in my opinion, the blur should take acount of the Z information, in order to know what to blur “on” and what to blur “behind”.
An arm passing fast “over” a chest should be blured completely over the chest, and not blured inside the arm boundary and sharp at the arm boundary.
in the same way if the arm is passing fast “behind” the chest, what we see of the arm should be blured but this should stop at the chest boundary, which is composited OVER the blured arm (and is static, so not blured).
For the moment, we have some blured part, but sharp edge at the discontinuity of motion…
-
-> For now you sometimes have to split your render into different passes
:: yep but it’s pretty complicated when you have a complexe geometry in movement,**** I have seen some people use Object Id tags (how to vary per render). It’s tricky due to their non anti-aliased nature and would deserve a tutorial. To handle the background case then (hole where the FG is), one can use RE:Fill otherwise the combination of bg and FG mblur might not total 100% opacity at some pixels.
-> In some cases (eg the background does not move), you can simply key out zero motion (0.5, 0.5, 0) off the Velocity image to avoid edge artifacts.
:: when i do this, it looks like if my character is motion blured, but then cutted out through it’s non blured alpha. in my opinion, the blur should take acount of the Z information, in order to know what to blur “on” and what to blur “behind”.**** This will also fail sometimes. Hopefully in some future we will address this better (with more UI to set and longer processing as a result). You can sometimes get better result from the optical flow based plugin for these.
Not what you want to hear but you can sometimes use one to patch the other with a quick very rough very feathered matte. Also the optical flow one allows you to select FROM or TO so you can do the inverse motion of the CG rendered flow, and then just blend the two results maybe.:: For the moment, we have some blured part, but sharp edge at the discontinuity of motion…
**** If you use the FG matte for FG pass on the MV and FG, and the inverse matte on BG for MV, you should not see some silhouette cutout edge (you might still see some intra-object motion blur anomalies). If so try choking the FG matte by 1.5 pixels to see.
Reply to this Discussion! Login or Sign Up