regarding velocity pass

[vance3d]

2 things.

I'm rendering out sequences with the velocity render element for use in nuke. does anyone know of any advantages to using real smart motion blur in nuke as opposed to using nukes built in vector blur?

the second thing is if I have to generate a velocity pass that rsmb can use I am figuring I have to use the method of setting a max velocity and clamping the result to generate the r and g information that real smart likes. is there any way of sampling the velocity without rendering? i thought i might be able to do a low res prerender and store the results of the "max velocity in last frame" entry in the velocity element and then determine what was the max for the whole sequence.

Thanks,

V Miller

[joconnell]

Heya Vance,

The only advantage of reelsmart is the fact that it can do it's own pixel tracking to make motion vectors for footage that don;t have their own. With your rendered velocity element, it's an accurate and pure source for the direction each pixel is moving as opposed to reelsmart which is doing some footage tracking and making a rougher guess.

On the second point since it's using the geometry to get the velocity information, you can turn off all lights and gi in your scene, and even sometimes apply a flat override material to everything so the rgb is as quick as possible - once you render the same animation from the same camera it'll be the exact same velocity info. I don't know if there's any smart way to get that max value without doing a render either.

[vance3d]

thanks for the answer Joconnell. In don't thing the compositors are planning on using anything but the vector pass I provide them for use with rsmb in nuke. I guess my question boils down to whether using vectors with rsmb in nuke has any quality or speed advantages over nukes own velocity blur.

[joconnell]

I wouldn't think there'd be much advantage as regards speed or end quality, the only thing that looks in any way interesting is that they've got some kind of background / foreground matte which might let you deal better with fast moving objects from smearing the slow or non moving objects behind them. They both seem to be multi - threaded and both support floating point values too so there's no major convenience or accuracy issue either. On that note, if RSMB can use float values in it's vector pass, then you shouldn't have to worry about looking at the max velocity per frame values either - a totally unclamped vector pass should do all the work for you.

[vance3d]

I was under the impression that rsmb wanted rgb info and the vray velocity render element is xyz data. If this is the case is there a method of converting the vray velocity data into something rsmb likes or does it basically have to be rerendered using the clamped method and setting a max velocity value?

[joconnell]

The vray pass just uses rgb data it seems to me when used at it's default settings - it just uses horizontal and vertical blur vectors (nuke only uses this, calls it u and v channels) and I think vray has the option to ignore movement towards camera (the z component). Likewise RSMB couldn't really track movement towards camera, only the left and right parts of it. I've only experience of RSMB in after effects and it didn't have many options to swap bits around, the nuke version might be better!

[instinct]

Note that i got no experience using RSMB, but there is not really a difference between xyz or rgb data besides the name. There might be a difference in the range that RSMB expects (most likely, as it prolly is not looking for signed data).

VRay Velocity (when used with the default settings) is outputting the pixel's movement as a vector of pixels per frame. It is a signed vector meaning that negative values mean movement to the left, positive movement to the right.

I would suspect that RSMB expects 0-1 values only. Prolly remapped as 0-0.5 beeing maximum left movement to not moving and 0.5-1 beeing not moving to maximum right movement. If this is the case then you need to remap your values from the max velocity range to that range. You can easily do that in nuke by setting whitepoint and blackpoint, or using an expression.

You might want to search the forums, i think there was quite some discussion related to rsmb, so there is probably better and more reliable information out there.

Regards,
Thorsten

[vance3d]

Thanks, I'll pass that on to the comp guys.