SEGI (Fully Dynamic Global Illumination)

Uhm, current Unity GI does not "run" at all, you need to compute it before playing.

 

This is interesting

@SebLagarde could you improve SEGI with this and bring into Unity 2018.3

 

That's HxGI by @Lexie

 

I have to say that some sort of VXGI solution for HDRP is a must at this point...

 

Armory has good voxel based GI https://armory3d.org/manual/#/graphics/global_illumination

Maybe something similar could be in HDRP.

 

Please someone make it to work with the Unity 2018.2.1f1

I get this error.
Thread group size must be above zero
UnityEngine.ComputeShaderispatch(Int32, Int32, Int32, Int32)
SEGI:OnPreRender() (at Assets/SEGI/SEGI.cs:979)

 

Weird it works when Voxel resolution is high but when I set it to Low resolution I get error! Anybody know why?

 

How is performance when you set Voxel resolution is high?

 

Performance is bad when I set Voxel Resolution to high. But with Low resolution it's a lot better.

 

Am I the only one having issue with Skinned mesh shadows ? I just tried to put Ethan inside my test scene and the shadow is just ... like big circle flickering around is feet, the size of the circle depending of the voxel space size. The only way to have a decent shadow is going as low as 7 (so the box is reaaaally small) on this gif the result is with the value at 30

 

You're not. I tried it a while ago and got all sorts of weird shadow. I don't know how to solve it.

 

You can put character to a layer that SEGI didn't see, attach an capsule to the character prefab, put this capsule to SEGI-only visible layer and receive an character shadow approximation from it.

 

Is this still being developed?

 

Well its open source but the creator dropped it like a hot potato. I feel like nvidia's big push for raytracing and the prospect of raytraced gi as well as the upcoming hdrp means this project is just dead in the water.

 

they will complementary, raytracing won't solve everything

 

at the core I believe nvidia's RTX & directx' DXR provide an api for BVH traversal and ray-triangle intersections, both of which could be used in a global illumination implementation but not so much a voxel based one.

I guess using the hardware accelerated intersections/traversal and coarse approximate geometry would make GI at interactive framerate work. It sounds more reasonable then raytracing reflections of high poly things on mirror things.

 

Think of it like it, you could kick start non the voxel gi by sampling using a precompute of the starting ray with the raytrace. Especially for sparse voxel.

Voxel have the adventage of low spatial frequency, ie it's better than noisy ray, so once you figure out all your ray intersection, you end with the voxel structure. Using the raytrace as a skip ahead of incoherent data toward coherent data.

It would also help with light leak, by helping voxel sorting what they are bad at approximating: geometry.

 

What is the difference between voxel illumination and raytracing ?

 

The support format and the type of result,
- voxel tend to be blurry (interpolated sample), raytrace tend to be noisy (single point).
- Voxel capture low frequency easily, raytrace capture high frequency.
- Voxel are an array (volume image), raytrace tend to use tree (bhv).
- Raytracing does analytic intersection of primitive, voxel use raymarching in the array.

 

Can anyone tell me what's the last version of Unity that SEGI works with? I just tried it on 2018.2.2f1 and it doesn't work.

Thanks.

 

My fork and CK's fork it's forked from both work.

https://github.com/ninlilizi/SEGI

Also, my fork has experimental support for Single Pass Stereo (Shameless plug)

 

Can SEGI be merged with progressive lightmapper as unified GI solution ?

Thank you @Ninlilizi for continuation of this project.

 

It won't, they don't solve the same problem at all.

 

I tried the CK fork of SEGI in Unity 2018.2.7. It works for new projects. However, when I brought in a project that was created in 2017.1 I get an error "Please fix compile errors before creating new script components". I have no idea what to do. Can anyone offer some guidance.

Thanks!

 

if you've compile errors... they usually appear in the console with a little red warning sign above the error telling you to fix the error..... and clearing the console doesn't clear out compile errors... So they stay there till you fix them. Click clear and see what remains... That's what the problem is.

If you can paste the stacktrace from the errors above said fix the errors. I'll advise

In other news, my fork now works in non-VR mode again as of a few hours ago. So feel free to try that too. I've been quite busy at it over the last week and intend to maintain it longer term as I'm working on a slow burn procedural world thing that I expect to take me a few more years to complete.

 

What are the differences between sonicether's last SEGI build and your current custom SEGI build - apart from the added Single Pass Stereo stuff?

 

Right now... CK's focuses on 2018 support and performance tweaks.... The volumetric shadow cam thing is a good example

Mine right now primarily adds support for SPS... That's my own primary use case... But going forward I'm going to be focusing on performance and extra configurability... Because I enjoy a challenge. Sonic very much cared about fidelity, and was reluctant to sacrifice it for performance. I plan to go in a different direction there. Maintain the shiny mode SE perfected, but add in some deadrekoned options. My own artistic style will bias me towards stuff that works for more styalized art than photorealism.

First couple of things added... An enhanced GI subsampling slider, that when at level 3 and 4 also substitutes the bilateral blur for a gaussian filter, as that gives a more aesthetically pleasing look at higher levels of subsampling. The trade off here is the texture sampling loop in the raytracer is a major performance penalty in testing. So that's helps when your metal is against the wall.. To compliment this, I've also added a tuner for the tracers noise coherency.. As the highfrequency noise that is acceptable to look at when running 1:1 gets really ugly and shimmery once you start subsampling... Turning that all the way gives blockier results. But if your not going for photorealism and instead more low-poly, flatshaded or similar stylized options. It actually kinda works, aesthetically speaking. Also, turning the noise filter all the way down into flat shading style output also gives a modest performance bonus. I'm using Voxel Farm in my project for my world building... So for a reference, that's my motivation... I've got a couple of profiles as examples.... The standard PC High quality settings..... And a 'VR' profile, which aside from the VR implication, gives is an example of pleasing results while taking advantage of these extra options I've added.

Edit:
As of now I have working reflections/specular tracing which the CK fork does not.

 

Here's an example of the current quality of the 4x downsampled GI mode:
This is with a GTX1060, and the scene has no lighting data.

On:
https://www.dropbox.com/s/h2k3vafomg3yswd/on.png?dl=0

Off:
https://www.dropbox.com/s/kgc31ht1l1ufhwg/off.png?dl=0

Here,s another example of a typical scene (from a random asset pack)

On:
https://www.dropbox.com/s/i08oiy44ktfwa1b/ON2.png?dl=0

Off:
https://www.dropbox.com/s/xpae4ddezao58y0/Off2.png?dl=0

Which shows we can still archive nice looking results at the bottom end of the fidelity/performance triage.

 

While we are at it, do you think segi's output can be cached using and onto texture rendering?

 

Depends on what you plan to do with the cached data.

 

@Ninlilizi Thanks for your help. Following your procedure, I was able to track down the culprit. Turns out it had nothing to do with SEGI.One nice thing about Unity 2018 is I'm getting 2 frames per second better performance than Unity 2017, for the identical scene.

@neoshaman Don't know if this will help you, but in Main Configuration if you uncheck 'Update GI' after you enter game mode, it's sort of like freezing SEGI's work. Caching, if you will. You will definitely get a bump in performance.

 

Does this include Forward rendering because it seems to be broken with SPS in Forward mode...

 

Still Deferred only at this point in time. I will look into adding some level of forward rendering support.

Edit:
Working on Forward rendering now... Just wragling with the specular mapping.

 

Horrah!

We now have Forward rendering support.... You loose some visual fidelity switching to forward. And the reflection tracing won't work with it... yet.... It's workable being the important thing ^_^

If your running in forward, give SPS/Instanced mode a go... It's speedy.

 

What are the advantages of using forward rendering for non SPS?

 

Forward rendering enables the use of MSAA Antialisasing... Which is a big deal to the VR crowd.

If your not doing VR, you don't need to run in Forward. The downside to it, is slightly lower visual fidelity in the resulting GI.

 

Forward rendering also supports object transparency/translucency.

 

My first stable release suitable for actual use is now ready https://github.com/ninlilizi/SEGI/releases/tag/v0.9.1

This release has feature complete support for both SPS and Forward rendering. Along with a pleathora of changes and optimizations.

There is limited support for SP/Instanced rendering... It 'works', but currently lacks a number of features.

 

Wow! That could make SEGI finally worth for VR development. Thanks a lot for the effort. Need to try it out ASAP!

 

Hi, i am surprised to see someone is doing a work on SEGI. What is the recommended Unity version to test it with !?

 

I'm using 2018.2

It 'should' also run on 2017.x ... But I haven't explicitly tested it there. If you encounter any issues, file them on my GitHub and I'd be more than happy to look into it.

Consider 2017.x the minimum I'm willing the support

Your welcome ^_^ .... Procedural VR use is my primary motivation. Now I've got a functional implementation for my next release I'll focus on performance.

 

I've had a request for a Demo scene. Something I'd not considered. So I will supply one when I get a moment.

 

Would be cool !

I tried it first in the new 2018.3.0 b1
- I had to import Post Process Stack and also switch from .NET 2.0 Subset to .NET 4.x Compatibility level to get rid of all errors.

Then i made a new project in 2018.2.6 ( only other 2018 i had ) and i only had to include post processing package withot needing to switch the .API Compatibility Level to .NET 4.x

However i couldn't make it run - i.e. getting proper GI.

So it would be cool to get some test scene also !

P.S. - also the camera in the scene coming with any newly created project has the "Default" tag instead of the "Main Camera" and i got a message to switch the camera tag to "Main Camera". However i still was getting the error after i switched the camera tag. Plus i got some Index Out Of Bounds errors but i am not sure now what it was about ! Will have to test again later !

Cheers !

 

Oh wow... This is embarassing..... My bad.... The bilateral blur function appears to be broken everywhere but my own test scene I've been using.

I'll fix that and push a new release with an included demo scene... Sorry guys

 

Thanks

 

All done. Bug fixed and 2 demo scenes added. I ported the original low-poly scene and threw together a sexier scene from the Sponza Atrium model.

https://github.com/ninlilizi/SEGI/releases/tag/v0.9.2

If anyone's wondering why there's currently a PostEffect Stack dependency. It's because I'm in the process adding SRP support (don't get excited yet. Got a learning curve and a massive refactoring ahead of me before that dream becomes a reality)

* Known Issues

SP/Instanced mode is still a work in progress. Adding Forward renderpath support was a complex problem to solve, as SEGI approximates a lot of stuff using data from the gbuffers, which are not a thing in forward so had to devise an alternate solution (Which I can and will improve on over time)... This new method of approximation appears to not work out the box for Instanced rendering, I need to further examine the situation there.

The toggle to visualise voxels doesn't work atm. That merely cosmetic as the voxelization is functioning normally.

Color saturation. It's lower than ideal. I plan to add some pre-process colour grading functionality to make this user tweak-able. This will also solve the next issue.

Lower contrast levels combined with setting the plethora of occlusion options cause the path tracer to not resolve the image. This results in breakdown of GI.

Performance... It's nowhere near as great as it could be. Got a fair idea of how to improve that and have an experimental branch running here where it's almost twice as fast as what this build achieves. So expect good things to come.

 

Thank you so much for your work @Ninlilizi !

Couple quick questions-

Known working and not working Unity Versions + OS? Good in 2018.2?

Still working in Deferred?

Thanks!

 

Oh my goodness, SRP support and performance gains? Where have you been all our lives @Ninlilizi?!? Thanks so much for your hard work and sharing it with the community.

 

Still works in Deferred. While I've massively refactored all the code over the last month... Most of that was to enable forward and sps.... Deferred rendering behaves mostly similar to before. Though there there have been some changes to the lighting attribution as I rolled part of the reflection pass into the diffuse path tracer as part of that process.

2018.2 is known good... I'm running Windows 10 with nvidia hardware and a rift for reference... I can't comment on vive support as those things are priced way out of my reach.

 

an error occured when use 2018.2.6
"
Assets/SEGI/SEGICascaded.cs(5,29): error CS0234: The type or namespace name `PostProcessing' does not exist in the namespace `UnityEngine.Rendering'. Are you missing an assembly reference?

"