Suggestion for a "little update" on the subdivision system.

[Douglas Silvestre]

First, sorry if this is the wrong place to post a suggestion, but I didn't find any other place on opensubdiv site to do it.

Second, I'm not a developer, so my suggestion is based on a idea that I have while I'm using Blender and "see a possible" performance improvement, that can reduce polygon count, what can increase overall performance in modeling, sculpting and reduce the amount of time in rendering. But again, I'm not a developer so technical information is not include.

The original post, with images: ADAPTATIVE SUBDVISIONS

The concept of the ideia is create a first step subdvision that can ignore a ffew faces, loops and divide the rest to create a more evenly distributed subdivsion, in this way, it's possible to reduce the not necessary divisions in small areas for example, like finger joints, eyes, nostrils and etc.

It's much more clear on the link with images.

The same concept can be used further like control the number of divisions in a entire scene at once for things like GEOMETRY NODES in Blender or used like a hibrid tesselation method, sending the data from the application, like Blender or similar software to a game engine that can use the information to reduce the polygon count to the lowest level or increase it to highest.

I can't post images in here, so, please take a look in the link.

[David G Yu]

Hi,

First, I would avoid calling this feature "adaptive subdivision" since that is overloaded with connotations and will lead to confusion.

It sounds like what you are suggesting is a modeling operation or modeling optimization tool which could be implemented using OpenSubdiv, but which itself is beyond the scope of what OpenSubdiv aims to provide.

What you are describing is a kind of mesh topology optimization, which indeed can be a very useful tool.

In order to optimize something like a character mesh you would need information beyond the mesh topology itself, e.g. nothing in the mesh topology itself identifies which parts of the mesh constitute a finger joint, so if you optimize the mesh for the character just in it's default pose you might optimize away important details that are needed to achieve a closed fist pose.

OpenSubdiv gives you a way to evaluate position, derivatives, etc. for any given pose, so you could build a tool that takes other information into account (e.g. a range of multiple poses, additional annotation of the mesh topology to indicate which regions of the mesh are affected by animation deformations, etc.) to identify candidate spans that do not have significant influence on the shape of desired poses and might be candidates to be removed, etc.

For background, I would suggest reviewing OpenSubdiv's overview of Subdivision Surfaces specifically the remarks about subdivision modeling operations vs the limit surface and the discussion of the distinction between mesh data and mesh topology.

Thanks!

-David