I think it would be great to have rotation velocity attribute (lets call it “rv” for instance) and maybe scale velocity attribute (“sv”) to be used to motion blur rotation and scale changes of point-instanced geometry (like velocity attribute do for translations).
It is useful when you have simulated RBD for a lot of identical objects, and, to speedup rendering (ifd generation part), want to point-instance them without loosing rotation and scale motion blur.

Scale motion blur is less important, but sometimes it's may be needed, I suppose.
P.S. I'm not sure if shear velocity needed (just to complete all kind of possible object transforms)

Still no reaction… Isn't that thing that useful?

Geometry is instanced onto points in the same manner as the Copy SOP. With the Copy SOP and Point Instancing you can use the following attributes to help orient and scale your geometry: N, up, pscale.

Ok, I've got your idea.
Actually I didn't knew deformation MB works for point instancing. (BTW, it's unclear why “deformation”?? Isn't it transformation blur for each instanced object?)

My RFE was related to “Velocity Attribute Blur” mode, which doesn't take any N, UP and pscale animation into account.

Anyway, thanks for the trick, it's really helpful.

Actually I didn't knew deformation MB works for point instancing. (BTW, it's unclear why “deformation”?? Isn't it transformation blur for each instanced object?)
Because it works in “object space”.

hoknamahn
Actually I didn't knew deformation MB works for point instancing. (BTW, it's unclear why “deformation”?? Isn't it transformation blur for each instanced object?)
Because it works in “object space”.

Well, not really. Deformation supports the arbitrary deformation of a geometry over time. To achieve this, deformation blur cooks your geometry twice: once at shutter open, and once at shutter close, and places the two shapes into the IFD.

Velocity blur cooks your geometry only at shutter-open, and uses the v attribute to fake a deformation blur for shutter-close.

Transform blur is indeed only what you can achieve by blurring from one transformation matrix to another (yes, you can get pretty creative with just that approach).

So while the granularity of point instancing is limited to a transform blur for each instance, overall, it is a deformation blur because the instance points have time-varying attributes that are read at both shutter-open and shutter-close.

– Antoine

Antoine Durr
So while the granularity of point instancing is limited to a transform blur for each instance, overall, it is a deformation blur because the instance points have time-varying attributes that are read at both shutter-open and shutter-close.

Thanks, it's much clearer now.