Precomputed Culling

Occlusion culling is the process of removing objects from the render pipeline that are not visible to the camera. Out of the box, Unity provides simple frustum testing algorithm. It is done by testing the object's bounding volume against the frustum of the camera. If the object is not within the camera's frustum, it is considered to be occluded and is not rendered.

Frustum culling is a technique used to cull objects that are not visible from the current viewpoint. Precomputed part of the occlusion culling suggests that the processing of culling is done before the game is built for final release. This principle is leveraging an idea that developers will have a lot of time and processing power to generate occlusion data, and reduces the complexity of detecting what is occluded in the runtime.

Frustum culling does not remove from the rendering objects that are behind of other objects.

Precomputed culling should be used for virtualy any 3D project. From fully open worlds, to closed, corridor based scenes. Scene with hundreds or thousands of objects usually without any culling algorithm can have significant GPU impact. Using precomputed culling can save a lot of GPU resources.

For open world games, in long view distant scenes you can save between 10-25% of GPU usage ,depending on the scenery and occlusion level.

For indor levels, you can save up to 90% of the GPU usage, depending on the scenery and occlusion level.

Culling data while using precomputed culling should be baked for a few main reasons. Before EVERY release.Scene changed significantly and artifacts are showing up during the development (and disabling culling fordevelopment is not an option).You want to test performance of your sceneYou want to test occlusion bugsWe recommend setting up CI pipeline that would bake culling after any scene changes to always keep the latestdatain the build, to not forget to run the pipeline. Baking everything before the release might not be the bestoption as bakes can last everywhere from few seconds to few hours.

We created this asset for 3 main reasons.

Umbra doesn't support multi-scene setups, or addative scene setups.
Umbra bakes are low quality and bleed a lot of objects.
We often had issues with baking in batch mode.

We will start by creating a new project, but you can also use your existing project.

New Project
In the Unity Hub, click on "Create New Project",
Select "3D Sample Scene (URP)" and give it a name,
Click Create project.

Open package manager, by pressing Window > Package Manager,In the top dropdown change Package to My Assets,Select Precomputed Culling and press Install,Done.

Locate the PrecomputedCullingProg.unitypackage package in file explorer.Double click on the package,In the new opened window press ImportDone.

Open the main scene where you want to initialize your setup,In the top menu, press Tools->Precomputed Culling->Setup,Optional - Create new GameObject, and add PrecomputedArea component to it.Adjust Precomputed Volume/Area to match the play area of your player (note you can have multiple areas),Adjust cell density and camera control (Find the best setting between bake Quality and Bake time and performance).Start baking either in Tools->Precomputed Culling->Bake or in AreaWait for bake to finish
Enjoy an increase in performance

For more advanced usages visit the related section in the repository

Open world areas:
Chunk size should be around 5 meters in width by 5 meters height by top jump height.
At least 3 spots in the middle. At top of jump height; At eye level; At Crouch eye level (add the fourth one if the prone position is allowed)
Allow for taking data from neighboring cells at distance of 1 reduce aggressiveness of the algorithm
Set runtime pixel requirement to 50-100 pixels
Allow for baking terrain
Allow for sprites baking (If the environment is not a pure 3D model)

Small Areas (building):
Chunk size should be around 1 meter by 1 meter by the height of the building
At least 18 spots. All spread out within the cell. At top of jump height; At eye level; At Crouch eye level.
(Optional if experiencing artifacts) Allow for taking data from neighboring cells in distance of 1 reduce aggressiveness of the algorithm
Set runtime pixel requirement to 0 - 25 pixels
Allow for baking transparent between 1 or 2 layers
More dense areas should have a bigger priority than less dense areas.

Support for URP
Support for multiple cameras
Support for multiple scenes with multiple variants and dynamic loading and unloading
Lightweight with an option for additional compression
Supports transparency
Supports LODs
Supports Unity Terrain
Supports 2D Sprites*
Runtime precision control between quality and performance
Support for all platforms including mobile, VR, AR, and WebGL
API for custom building pipelines
Everything is written in C#, full source code ncluded
Code documentation + external documentation with tutorials
Easy to setup
Examples and Demos
Support via email or discord

Required to bake scenes and their variants before build.

Support for Legacy (built-in) and HDRP
Support for dynamic objects
Support for occlusion portals
Split baking between multiple computers
Improve baking times
Better debugging tools and statistics
Dynamic lighting shadow