Workbench-Next Volumes Improvement #104642

New Issue

Clément Foucault · 2023-02-12T00:58:30+01:00

Clément Foucault commented

2023-02-12 00:58:30 +01:00

A deep rewrite of volumes rendering system could bring several improvement. Here are some goals.

Half/Quarter resolution rendering

This would improve performance for really dense volumes or Hi-DPI displays.
An up-sampling method should be used for nicer intersection with the scene geometry.

Improved lighting & shadowing.

Shadowing is currently non-existent and lighting is very flat.
The lighting should follow the lighting used by the surfaces (orientation, color) in all lighting mode (studio, matcap, flat).
For matcap, we need a way to extract lighting as light.

Here is a possible algorithm to achieve good shadowing:
For each ray step into the volume, shoot one shadow ray towards the most powerful light, and one shadow ray towards a random direction to get ambient light shadowing. The emission texture could also be sampled along the way to produce self illumination.

All of this is going to be expensive so it should be optional. Hopefully, lower resolution rendering should help.

Correct Clipping support

Support for the Alt+B clipping is currently missing. It should be supported in the new implementation and allow better inspection of volume internals.

Clipped volumes should be lit as if not-clipped.

Depth writing for correct intersections with overlays.

Volumes currently do not write to the depth buffer producing no intersection with the floor grid and other overlay. This is distracting for depth perception.
The resolve (up-sampling) pass should write to the depth buffer.

Material roughness could affect anisotropy of scattering.

Default anisotropy is not suited to all visuals. Being able to tweak this parameter could help work with different simulation types.

Different volume intersection

Volumes are currently rendered back to front with ordering issues if they intersect. This problem seems difficult to fix while tackling the other task and keep the rendering cost manageable.

A deep rewrite of volumes rendering system could bring several improvement. Here are some goals. ### Half/Quarter resolution rendering This would improve performance for really dense volumes or Hi-DPI displays. An up-sampling method should be used for nicer intersection with the scene geometry. ### Improved lighting & shadowing. Shadowing is currently non-existent and lighting is very flat. The lighting should follow the lighting used by the surfaces (orientation, color) in all lighting mode (studio, matcap, flat). For matcap, we need a way to extract lighting as light. Here is a possible algorithm to achieve good shadowing: For each ray step into the volume, shoot one shadow ray towards the most powerful light, and one shadow ray towards a random direction to get ambient light shadowing. The emission texture could also be sampled along the way to produce self illumination. All of this is going to be expensive so it should be optional. Hopefully, lower resolution rendering should help. ### Correct Clipping support Support for the Alt+B clipping is currently missing. It should be supported in the new implementation and allow better inspection of volume internals. Clipped volumes should be lit as if not-clipped. ### Depth writing for correct intersections with overlays. Volumes currently do not write to the depth buffer producing no intersection with the floor grid and other overlay. This is distracting for depth perception. The resolve (up-sampling) pass should write to the depth buffer. ### Material roughness could affect anisotropy of scattering. Default anisotropy is not suited to all visuals. Being able to tweak this parameter could help work with different simulation types. ### Different volume intersection Volumes are currently rendered back to front with ordering issues if they intersect. This problem seems difficult to fix while tackling the other task and keep the rendering cost manageable.

Clément Foucault added the

Type

To Do

label 2023-02-12 00:58:30 +01:00

Clément Foucault added this to the EEVEE & Viewport project 2023-02-12 00:58:31 +01:00

Miguel Pozo referenced this issue

2023-03-03 13:44:41 +01:00

Workbench-Next #101619

Clément Foucault added the