Do you have any example renders to demonstrate the improvements? Hole filling was always a big problem with the existing node. Is the algorithm similar to Pixmotor?

Hopefully this code can replace the existing vector motion blur, ideally there's just a single node for this purpose.

For the Cycles changes, I see you made motion vectors unantialiased, I guess you found this works better for your algorithm. Is there a specific reason you also removed antialiasing from the UV and normal passes?

I'll have to track down some production renders for you. The node is actually 2 algorithms. Fat mode doesn't require any hole filling because the transparent pixels always ramp up from the object borders which sometimes makes objects look a bit fatter (hence the name). This also makes it suitable for use over frames rendered with a single layer. The other (non-fat) mode is really only suitable for objects that are rendered in their own layers. It doesn't make the objects look fatter but it is susceptible to alpha holes. The hole filling algorithm sounds similar to what Pixmotor describes from what I can tell. An estimate of the alphas for the holes are made and compared with the unblurred object alpha. Then the color of the hole is "boosted" to fill it in.

As far as the vectors go, we were finding that AA vectors didn't make any sense for either of the objects overlapping that pixel. i.e. two objects moving perpendicularly made a vector that didn't match either. And since this node comps based on object ID the edges were shooting off in random directions. Ideally I'd love to have the vectors for the object that covers most of the pixel (like Mental Ray) but this was easier.

Any progress here? I ask, because this is a useful feature. And there seems to be no activity here.
@Sean Kennedy (hype) Adding you here too. :-)

Hi @Account Deleted (tod_baudais) any progress here?