Question What's the plan for URP reflections improvements?

Hi, I remember that there're plans to improve reflections in URP.

And in 2022.2, the reflection probe atlas was supported in Forward+ to improve probe blending.

So what's the next step? Or is it not a priority yet?

 

No Planar reflection?

 

not OP but planar reflections seems like a must in any pipeline. For a water plane or for mirrors. Sure there is a cost and scalability needs covering but I do not think you can do without built-in support.
Not just games but for arch-viz (which HDRP may seem better for but that has a high entry point).

 

Thanks for the detailed reply!

Yes, I use URP because of mobile platforms support. I recommend prioritizing improvements to URP's reflection probes (rotation, sphere projection, runtime convolution?) first, because I believe this can only be integrated well by the URP team.

There're some planar or screen space reflection solutions for URP on the Internet and I think there'll be more.

 

Planar reflection probes seem to be expensive on mobile but it'd be nice to have, because it's the only (non ray-traced) solution that supports mirror-like reflections.

I saw that many mobile games implemented screen space planar reflection (a less accurate SSR without ray marching). I wonder if it's possible to implement this as screen space planar reflection probes (re-project color from previous frame) in URP as a faster solution compared to HDRP planar probes.

 

For me SSR is a must cause it's much more easily universal, then dynamic relighting for more dynamic scenes (ToD?), then planar reflections since that is the most circumstantial.

It's not like solutions for planar reflections don't exist, it's just they're a pain to use in my experience. SSR is annoying sometimes but at least it just works.

 

Agree that reflection probe relighting would be a useful feature for URP. This can be a cheaper way to have dynamic lighting when combining with dynamic APV.

But I'm not sure how this will be done in HDRP and if it's possible (or suitable) to support this feature in URP Forward and Deferred (no reflection probe atlas).

I think the problem is that the development of new features in URP is a little slow.

 

screen splace planar reflection using dx11 faster x5-10 times than any other SSR method and have pixel-perfect reflection.
https://github.com/ColinLeung-NiloCat/UnityURP-MobileScreenSpacePlanarReflection
https://remi-genin.github.io/posts/screen-space-planar-reflections-in-ghost-recon-wildlands/

Here is comparison of my SSRP realization (with holes reprojection) vs HDRP reflection with high quality (48 rays)

Spoiler: Gif 5mb

 

Yes, It'd be very useful to many URP projects. But considering the use of atomic operations (as shown in the link), I think the URP team need to find another way to implement SSPR (when it's planned).

 

InterlockedMin with textures doesn't supported on metal (and partial support on android), but InterlockedMin with StructuredBuffer works perfectly.
As vulcan supported on 85%+, so it must work everywhere. (https://developer.android.com/about/dashboards)
I can share my code if it helps for unity devs.

 

How does this work when you look down? It seems to have similar occlusion issues to SSR, but it doesn't cause a huge stretching effect. Isn't the point of planar reflections is to see offscreen or occluded information?

 

SSR can work on planar surfaces too and will look the same as SSPR(if I understand correctly) so what's the difference? Performance?

 

It's screen space reflection, which works only for planar surfaces (like a water). It's not a planar reflection (which requires the second camera and double rendering)

 

More performant and fewer artifacts.

 

I think the control of SSPR can be terrible if the solution simply applies to a shader and attaches a script to an object.

It would be more useful to implement SSPR as a mode of the planar probes (screen space, additional camera) in URP. This also makes rough reflections possible with SSPR.

 

I'm not sure if it's a common issue of SSPR, but it seems that in scenes with complex occlusions of (foreground & background) objects, SSPR produces more artifacts than (well-designed) SSR.

(sorry that I cannot find a screenshot to show this)

Regardless, it should still be a performant solution worth considering for mobiles.

 

It's not possible, because SSRP reflect all depth pixels in "reflection buffer", while any other SSR trying to find this "depth pixel" using raycasting with 100500 misses and artifacts.