I'm trying to make sense out of Point Deform SOP. For example:



The left is a very high-poly mesh. The middle is its low-poly counterpart. The right is the deformed low-poly.

Now I connect these three to Point Deform SOP. I'd expect something like this:



However, instead, I got this:



I played with all Point Deform's parameters. No luck.

I know in this simplified case, Lattice or regular bone deformation may be a better choice. But I was reading the document of H19's Muscle and Tissue System [www.sidefx.com]. It uses Point Deform to transfer the simulated result to the high-poly mesh. (It doesn't even use Delta Mush. Just a plain simple Point Deform!)




If I couldn't even make this simplified case work, I can't imagine how people use Point Deform to deform something as complex as human body. So I assume there is something that I missed and it's actually possible to get a smooth, nice result out of Point Deform.

raincole
If I couldn't even make this simplified case work, I can't imagine how people use Point Deform to deform something as complex as human body. So I assume there is something that I missed and it's actually possible to get a smooth, nice result out of Point Deform.

Point deform is perfectly capable of nice results, but you can't have a bind pose with overlaps. It captures based on proximity only, so if topologically distal points are proximal in the capture pose, then the resulting bind will be garbage. You want your bind pose to be more like a straight tube--no overlaps. This is the reason nobody rigs a character in a natural pose--it's done with a t-pose.

jsmack

raincole
If I couldn't even make this simplified case work, I can't imagine how people use Point Deform to deform something as complex as human body. So I assume there is something that I missed and it's actually possible to get a smooth, nice result out of Point Deform.

Point deform is perfectly capable of nice results, but you can't have a bind pose with overlaps. It captures based on proximity only, so if topologically distal points are proximal in the capture pose, then the resulting bind will be garbage. You want your bind pose to be more like a straight tube--no overlaps. This is the reason nobody rigs a character in a natural pose--it's done with a t-pose.

Thank you. I've realized it. I really wonder if anyone made a "topologically-aware point deform" tho. For example, parts like fingers and lips are often very close to each other, and proximity is not a good way to bind them.

I noticed that one could modify the weight captured by Point Deform, and made my own version. It's not perfect, of course. If anyone likes to try/improve it, I've attached my HDA.

You might want to use a biharmonic capture with that setup. Might produce better results.
https://www.youtube.com/watch?v=miwEX8vR_2c&ab_channel=Houdini [www.youtube.com]