google/filament

Cutoff 0.25 test not passing for AlphaBlendModeTest.gltf

cx20 opened this issue · 12 comments

cx20 commented

Describe the bug
I tried to display AlphaBlendModeTest.gltf with filament latest version.
However, it did not pass the Cutoff 0.25 test.

To Reproduce
filament latest version + AlphaBlendModeTest.gltf result

Expected behavior
image

Screenshots
image

Desktop (please complete the following information):

  • OS: Windows 10 Version 1903 (OS Build 18362.295)
  • GPU: Intel HD Graphics 520
  • Backend: ANGLE
  • Browser: Chrome 76

This is expected when MSAA is enabled. The glTF spec is ambiguous with respect to ALPHA_TO_COVERAGE:

KhronosGroup/glTF-Sample-Models#221

cx20 commented

Thank you for telling me that thread. I will look at it.

This test now passes properly.

cx20 commented

@romainguy I tried the gltf-test sample again with the latest build of Filament. However, the Cutoff 0.25 test result was 0.50.

Filament latest version + AlphaBlendModeTest.gltf result:
image

Is it the wrong way to use it? It would be great to get some advice.

Are you using our main branch?

Screen Shot 2020-08-10 at 4 20 27 PM

cx20 commented

Are you using our main branch?

Yes. I used the following build of the main branch to try the WebGL version of Filament.

image

@prideout Any idea? My screenshot is from gltf_viewer (native, not WebGL)

There's nothing to do here on the Filament side. When MSAA is disabled, we "pass" the test. When it's enabled, the rendering is affected by how the browser and GPU implement ALPHA_TO_COVERAGE.

Note that other renderers may or may not enable ALPHA_TO_COVERAGE since this is unspecified in the glTF standard.

It looks like cx20 has MSAA enabled, and is using a browser or GPU that implements ALPHA_TO_COVERAGE strangely. Not a Filament bug.

Ah right, because when I disable MSAA I get the following in native:

Screen Shot 2020-08-10 at 5 02 04 PM

With MSAA:

Screen Shot 2020-08-10 at 5 02 39 PM

So yeah looks like a browser issue. I guess we could always disable alpha to coverage for glTF but ugh…

I don't think we should take any actions until Khronos resolves KhronosGroup/glTF#1668

Not holding my breath :)

This bug is 3 yrs old, closing.