Better rounded edges Tweet

[Andreas_Resch]

Maybe the devs can take a look at how to improve the rounded edges algorithm. It looks fine with small values but when they get a big higher, in some cases they fall apart and the results look really ugly. I did a comparison in Cycles and the algorithm is much better there. It's not something I need all the time, but then and now I need to get a bit closer to objects and then the issues really show.

I've attached a comparison.

And here's an OBJ file that shows the corner in the render test. A rounded edge value of 0.05 was used in both cases (Cycles and Octane).
https://www.argfx.at/upload/RE_Test_01.zip

[elsksa]

Which Round Edges configuration was set in Octane?

There is all these options:

[Andreas_Resch]

I've tested them all. Accurate is the one used in the image. "Fast" looks even worse.

[elsksa]

but then and now I need to get a bit closer to objects and then the issues really show.

Round Edge is only a handy fast and efficient workaround for the lack of mesh level beveling, such as triangulated CAD data. It is better to do it at mesh level if it's a quad poly, especially for detail-critical scenario.

But if it can be improved (if it needs to), it's a fair request.

[Andreas_Resch]

As shown in the Cycles version, there's room for improvement And when you triangulate STP objects, in most cases cleaning up the mesh and physically beveling them is out of the question. Nobody has the time (and down the line money) for that.

[elsksa]

As shown in the Cycles version, there's room for improvement And when you triangulate STP objects, in most cases cleaning up the mesh and physically beveling them is out of the question. Nobody has the time (and down the line money) for that.

I replied a second time prior to seeing your reply and indeed, that's a case where an iteration of the pre-existing improvements is fair and worth it. Your case joins what I previously mentioned with such atypical cases where Round Edge comes as a time saver, if not mandatory.

[elsksa]

I'd suggest sending it here, just in case it doesn't receive the attention it deserves.

[Andreas_Resch]

Thanks for the link. Let's wait a bit and see.