WIP: Use a different method for overlay edit mesh bias #104921

Jorijn de Graaf · 2023-02-18T14:46:50+01:00

Jorijn de Graaf commented

2023-02-18 14:46:50 +01:00

This is the same way I did it for retopology, and it works there.
Don't merge this, it's just a proof of concept.

This is the same way I did it for retopology, and it works there. Don't merge this, it's just a proof of concept.

Jorijn de Graaf added 1 commit 2023-02-18 14:47:02 +01:00

1d92979411 Use a different method for overlay edit mesh bias

This is the same way I did it for retopology, and it works there.

Jorijn de Graaf requested review from Clément Foucault 2023-02-18 16:43:38 +01:00

Clément Foucault commented

2023-02-22 13:47:50 +01:00

So, I now remember where this offset comes from.

It was originally using glPolygonOffset(1)(2) for that. I don't remember the values used at the time, but I remember the guideline was to get the same result with the new implementation. ED_view3d_polygon_offset was then used to emulate the glPolygonOffset when moving to GL 3.3 (3).

glPolygonOffset has 2 parameters:

factor, which gives you a bias equal to factor * Z_delta. This wasn't used in the previous implementation and isn't emulated in the current one for performance and accuracy reason (it can make edges poke trough surfaces that are parallels to the view plane).
units is basically a factor to an epsilon value chosen by the implementation. I remember vaguely that we only used 1 unit of bias.

This presentation (4) gives a good derivation to what the bias should be. The espilon value should be as low as possible in normal case and adjustable in the retopo case. Note that the slide 21 state Z_ndc = ... + epsilon. Since we are biasing in clip space this epsilon value becomes epsilon * W_clip. Note that W_clip = gl_Position.w in the code.
The abs(W_clip) comes from me not wanting to change the bias sign for (questionable) clarity.

There are confusion about NDC Space vs Clip Space coordinate throughout the codebase. I am to blame as I conflated both terms for a long time. I created #105070 for tracking this issue.

References:

So, I now remember where this offset comes from. It was originally using `glPolygonOffset`(1)(2) for that. I don't remember the values used at the time, but I remember the guideline was to get the same result with the new implementation. `ED_view3d_polygon_offset` was then used to emulate the `glPolygonOffset` when moving to GL 3.3 (3). `glPolygonOffset` has 2 parameters: - `factor`, which gives you a bias equal to `factor * Z_delta`. This wasn't used in the previous implementation and isn't emulated in the current one for performance and accuracy reason (it can make edges poke trough surfaces that are parallels to the view plane). - `units` is basically a factor to an epsilon value chosen by the implementation. I remember vaguely that we only used 1 unit of bias. This presentation (4) gives a good derivation to what the bias should be. The espilon value should be as low as possible in normal case and adjustable in the retopo case. Note that the slide 21 state `Z_ndc = ... + epsilon`. Since we are biasing in clip space this epsilon value becomes `epsilon * W_clip`. Note that `W_clip = gl_Position.w` in the code. The `abs(W_clip)` comes from me not wanting to change the bias sign for (questionable) clarity. There are confusion about NDC Space vs Clip Space coordinate throughout the codebase. I am to blame as I conflated both terms for a long time. I created #105070 for tracking this issue. References: 1. https://registry.khronos.org/OpenGL-Refpages/gl4/html/glPolygonOffset.xhtml 2. https://www.opengl.org/archives/resources/faq/technical/polygonoffset.htm 3. https://github.com/dfelinto/blender/blob/87a0770bb969ce37d9a41a04c1658ea09c63933a/source/blender/gpu/intern/gpu_matrix.cc#L690 4. http://www.terathon.com/gdc07_lengyel.pdf (p. 18 Depth Modification)

Jorijn de Graaf commented

2023-02-26 18:38:29 +01:00

I've gotten a response and talked about it further on the Blender Chat, and I've learned more about matrix math; basically this PR is no longer needed, so I'll close it now.
Here's the resources that helped me the most:

https://www.scratchapixel.com/lessons/3d-basic-rendering/perspective-and-orthographic-projection-matrix/projection-matrix-introduction.html (and subsequent pages)
http://www.terathon.com/gdc07_lengyel.pdf (pages 18-26)

I've gotten a response and talked about it further on the Blender Chat, and I've learned more about matrix math; basically this PR is no longer needed, so I'll close it now. Here's the resources that helped me the most: * https://www.scratchapixel.com/lessons/3d-basic-rendering/perspective-and-orthographic-projection-matrix/projection-matrix-introduction.html (and subsequent pages) * http://www.terathon.com/gdc07_lengyel.pdf (pages 18-26)

Jorijn de Graaf closed this pull request

2023-02-26 18:39:55 +01:00

Clément Foucault referenced this pull request

2023-02-26 22:04:31 +01:00

#70267 Retopology Overlay #104599

Jorijn de Graaf referenced this pull request

2023-02-26 22:37:00 +01:00

#70267 Retopology Overlay #104599

Jorijn de Graaf referenced this pull request

2023-02-26 22:53:43 +01:00

#70267 Retopology Overlay #104599