Antialiasing in headless rendering

[pglira]

Thank you for creating this amazing tool!

I've noticed that the visualization quality significantly degrades when using headless rendering mode. I suspect this is due to antialiasing not being applied in headless mode.

Here’s a comparison for reference (left = headless, right = original).

Is there any way to achieve comparable visual quality in headless mode?

For context, I’m using the Python bindings of Polyscope.

[nmwsharp]

This is interesting, I'm not sure quite what's causing it, it would be good to fix.

As you say, antialiasing is one possible cause, it is controlled explicitly with ps.set_SSAA_factor(2) (for instance). However, this shouldn't be changing between local vs. headless renders unless you are setting it intentionally.

Another possibility is that the points are different sizes, which is controlled by the radius parameter of the point cloud. (and might be a function of the global lengh_scale if you are using a relative radius. But again, that shouldn't be changing unless it is set explicitly.

Maybe there's a bug somewhere and one of those values is getting set differently for local vs. headless.

Looking closely at the screenshots, it looks like it's not just that points are smaller on their boundaries, but some points are not contributing to the image at all. It's possible that this is due to some very low-level openGL/driver/EGL behavior for rendering of quads which are smaller than 1 pixel... In that case it might be tricky to fix.