dmeville / volumetric-occlusion-mask-srp

Using SRP to create a volumetric occlusion mask for objects to use in their render pass

WANT TO BLOCK STUFF FROM GETTING WET?! I GOTCHU FAM

Edit 7/23/2018 - The system works okay as is, but I'm not really satisfied. Will be looking into make a system based around mesh distance fields to achieve this effect.

Using Unity 2018.2.0b7 Package Manager: Lightweight Render Pipeline 1.1.10-preview and Post-processing 2.0.7-preview

Most of the code is in LightweightPipeline.Render, be sure to replace the version of LightweightPipeline.cs that comes from the package manager with the included one in the project.

Occluder objects need a convex closed mesh renderer and an Occluder behaviour.

Access the global shader texture _CustomOcclusionMask to use the generated mask in an object's shader (to mask it's colour or whatever) or use the incldued CustomOcclusionMask.Asset Amplify Shader Function with ASE.

There is also an included ShowOcclusionDepth post process that shows the mask

Is probably slow and unoptimized, has some issues I'm still working out, but is a good starting point if someone is looking for a system like this.

---

https://www.youtube.com/watch?v=wDp5uLv2HrU

Been learning about the new scriptable render pipeline in unity 2018, stumbling along because I have no idea what I'm doing.

I wanted to make a system that would allow me to mask out shader effects based on other geometry and work properly with masking by depth. For example, in a snowy forest, I would like to control snow coverage with a global shader variable to affect all meshes, however some meshes (like inside a house) shouldn't get snow on them. I could manually paint masks for individual geometry but that sounds dumb.

Instead I create a cube (or any other concave closed mesh) and mark it with a special two pass shader. This shader renders the backface and frontface depth of the cube to calculate the cubes volume, and uses that to compare with the depth texture. If the pixel we're trying has a depth that falls between the frontface and backface depths, we know that pixel is within the blocking volume and should be masked out.

Volumes are baked into a OcclusionMask texture which I can sample in any shader to access the mask.

I will use this to easily make it so shader wet-ness effects (like animated water dripping on rocks) or snow on the ground doesn't appear in places they shouldnt (like inside houses, caves, etc). What's cool is that it obeys depth properly, so if we're looking "through" a occlusion mesh, only the pixels within the volume are masked, pixels "behind" the mask but off in the distance are not affected (as they would be if it was a simple screen space effect!)

More info of me trying to work through this problem on reddit: https://www.reddit.com/r/Unity3D/comments/8f2stm/how_would_you_go_about_making_a_shader_that_color/?st=jidwwnr0&sh=c9ff5f64

---

Do whatever you want with this code. Have questions?

• Twitter: @DMeville
• Email: [email protected]

volumetric-occlusion-mask-srp's Issues

Only one blocker is rendered to the mask

Not sure when this broke, but it was working in the first video... and the gifs..

Performance/Occlusion Component

Needs testing. Currently all scene occluders are rendered, and are NOT culled by camera view or anything. We also keep our own list of occluders, surely we could get this list and iterate through it in the SRP directly. Not sure how to fix this.

https://forum.unity.com/threads/feedback-wanted-scriptable-render-pipelines.470095/page-14#post-3532299

Too Many Passes?

Could probably combine the backface and frontface depth passes into a single pass (cutting down on half the rendertexture swaps and clears).

We can get the sign of the face in the shader with the vface param, there's some sorting issues though that I need to work through because SHADERS ARE HARD and I don't know what I'm doing.

Should be able to render the backface depth to rt.r, frontface to rt.g, backface mask to rt.b, and frontface mask to rt.a. I think. Would be way better

Depth Issues

If we stick the camera so that only backface is rendered to the temp mask, things break because it doesn't have a corresponding frontface to test against so the mask disappears.

Mask Softness

Right now it's only sharp edges [0 or 1] values in the mask. It would be nice to have a "softness" falloff somehow. Can't really do thickness [frontface - backface depth] because that results in an object getting "thinner" the farther away from the camera it is. Precision issues or non-linear depth values? idk

Could maybe reconstruct world position from depth, and use that and the distance to the (closest?) face? not sure.

Shadows broke

dunno, probably an issue with one of my custom shaders