Official New Post-processing Stack (Pre-Release)

Any plans to re-introduce the missing reinhard tonemapping? I think for outdoors it's results were better than ACES or the neutral curve one.

 

Hey guys... now that 5.5 is out, I have a question to regards to scripting this via runtime. I am using the UFE asset and since the levels are just prefabs, I am looking for a way to script the prefabs so that the camera (Which is also a prefab called "LoadCamera" and that is used for all of the levels) can use the PostProcessingProfile I have.

Example, when my level "The Sanctuary" loads, the script will find the load camera, and assign the Profile to the camera I want.

 

Hey,

do you guys have actual numbers on how better the performance is?

Cheers,

 

I don't think any generic "numbers" can possibly be used as a comparison: it's an effect stack - so the performance benefit increases the greater the number of separate effects you replace. It's not particularly that the DoF effect in the stack is better performing than the standalone DoF component (though it may, or may not be).

 

I can say for sure than the SSR inside the stack is performing incredibly well (maybe 2x-3x better? correct me if I'm wrong, but personally I thought we'd never have that effect with the old cinematic effects). The only thing lacking that I had to fix (took me a few mins) was the lack of Reinhard tonemapping that was scaling incredibly well with my outdoor island environment.

But I would strongly recommend the stack.

 

Running even a fairly lean PP stack for prototyping purposes (to try and better profile asset speed) it seemed to shave just about 0.5ms per frame off, @ 720p and on a GTX1070. Bear in mind both the previous and new setup aren't optimized other than turning things on.

It looks like the general performance is pretty good. Haven't given SSR a decent thrash yet, but AA, tonemapping, DoF and AO are looking great and running fast. Also, the motion blur, at higher framerates, looks fantastic.

 

AO isn't fast, SSR isn't fast. Haven't tested grading/tone and DoF yet, but I'm pretty sure the AO is slow, have to use the bare minimum settings on that one

 

Hi guys, I have a question regarding the stack: how can I access its individual components (bloom, DOF, etc...) via code to change them in real-time (when the game's running)?

 

I don't even use AO or SSR... if you have AO in your textures, you don't really need it... it is just slowing everything down

 

Well without nearby "contact" shadows, it's kinda needed yeah, especially as it's possible for AO to be really, really fast - 0.5ms on PS4 if I'm not mistaken with GTAO. Particularly impressed how close to ground truth it is, and it even has specular occlusion.

 

man i really hope that we getting a specular occlusion soon, speaking of performance, ssr is mostly the one which affect performance from my test.

 

And even on my laptop, it is pretty quick... is there any scripting reference on adding a Profile at runtime? I want to add a script to an empty object to just assign a Profile of my choosing. (I'm finally learning to script a bit)

 

Well, if it's not fast in general, it is, for sure, 'faster' than the Cinematic effects one. At least in my scene's case.

 

It's the same one though, and it's still 100% unusable on console. It needs to be made considerably faster (in our case).

 

I'm sorry, I realize maybe my question maybe wasn't clear: How do you modify an effect in real-time, using C#?

Let's say you need to change bloom intensity. Currently we do:

Code (csharp):

1. myCamera.GetComponent<Bloom>().bloomIntensity = 1f;

But now that the effects are driven by a profile asset which lives in the project, how do you access the individual effects?

Also, if you make changes to the profile asset during gameplay, do those changes remain as is, or are they reset once you exit the game (in the editor)?

 

@Chman @Panokani
Can you please take a look at the questions about tonemapping I've posted earlier?
https://forum.unity3d.com/threads/new-post-processing-stack-pre-release.435581/page-3#post-2872342

You can get a reference to the profile from the behaviour component and then and modify fields on it. If you don't want to modify a profile object (you probably don't, since they will likely be used as quality presets or something), the best approach is to copy the profile object before assigning it to the camera, and perform changes on that copy. I think it was discussed in detail a few pages ago.

 

Thanks Thomas! I'll take a look at that.

 

Happens on builds as well, as the build is "the first run" also...

 

Good morning @Chman, is there a script reference for this? I am trying to do a setup for my prefabs in UFE to minimise set up.

 

This will be applied to a prefab... if it's anything similar to assigning a skybox to a scene at runtime, should be easy to do.

 

Seems to break on latest 5.5p1 released today with pink screen:
Invalid pass number (2) for Graphics.Blit (Material "(Unknown material)" with 1 passes)

This all smells like serialization issues. I removed the pp stack and re-added it, same error, then restarted Unity and it's fine.

NullReferenceException: Object reference not set to an instance of an object
UnityEngine.PostProcessing.PostProcessingBehaviour.OnPreCull () (at Assets/PostProcessing/Runtime/PostProcessingBehaviour.cs:104)

Still exists, but goes away when I disable InControl input Manager.... fecking odd considering all was well in pre-f3 land.

 

Any update on this?

 

Any plans to support Luts for color correction?

 

Code (CSharp):

1. using System.Collections;
2. using System.Collections.Generic;
3. using UnityEngine;
4.  
5. public class SetPostFXProfile : MonoBehaviour {
6.     public Profile postfxprofile;
7.  
8.     // Use this for initialization
9.     void Start () {
10.         LoadCamera.GetComponent<PostProcessingBehaviour>().Profile = postfxprofile
11.     }
12.    
13.     // Update is called once per frame
14.     void Update () {
15.        
16.     }
17. }
18.  

So here is some code I threw together... I basically want the script to find my LoadCamera, and just assign a profile... I am not a programmer so I am sure a lot of this syntax is not correct...

 

I'm really enjoying this combined stack, the ability to tweak things in play mode easily is a giant help when it comes to tweaking content.

Of course it is a bit easy to abuse the effects when you use too many at once. Earlier today I was trying to recreate the ReCore look (using realtime GI with a simplified environment mesh in a layer culled from the camera view providing the GI to the light probes) and I quite enjoyed the results when I used this processing stack on top.

 

I've been trying to figure out how to get GI working properly for my outdoor levels... I would love to know the workflow to cull the mesh from the camera for light probes... i love the red tinge from the dirt, something I am having a hard time getting correct in my lighting...

 

It's just a basic mesh as low poly as possible that roughly matches the more detailed mesh placement. The basic mesh is the only thing marked as lightmap static, with only Realtime GI calculation turned on. It's then placed on a layer of its own which is simply culled in the camera settings.

We probably shouldn't get this particular thread too off-topic considering it's about the post effects stack, I'll start a thread for my workflow if it turns out to actually be useful performance wise once I've done some proper comparisons and profiling on it.

 

Or you can pm me... But on topic, this new Post Effects stack is the closest I have gotten to film quality

 

Feature Request: Screenspace shadows like in here:

https://forum.unity3d.com/threads/wip-se-screen-space-shadows.441909/

Why?

1. native solutions are best
2. it'll be faster
3. it's far, far more impressive than AO while being cheaper and more usable... don't need character shadows like in blacksmith when this is faster and better.

 

Agree it would be nice to have. I second the feature request. However, I very very strongly disagree that its more important than SSAO. I find your opinion on that very surprising.

 

SSAO is overrated and highly unrealistic. Corner shadows should come from the lightmapper, so no need for SSAO there. The only thing that it's useful for, is some fake indirect shadowing for dynamic objects. But there are better solutions for that (look at naughty dog papers for the last of us).

And all in all, I don't think the performance hit is worth the benefit of grounding the characters somewhat, while also making all the corners in your game look like they have mold.

 

Because I doubt I have the perf budget for both. What would be nice is some new invention that combines both techniques for SSAO and SS shadow

Go on Unity, you've been innovating so far, why stop here?

Basically, if it means I'm allowed to reduce my main shadowmap rendering requirements, it might even go a long way to breaking even because SSAO and SS Shadows do not require re-rendering the scene. That's a lot more quality for similar bucks in theory, assuming they can be optimised together.

Also, one cannot rely on 3rd party due to 3rd party hitting limitations of the Unity API.

 

I'm not sure if it's me but, inside the HTC vive I can see the SSRR as being slightly off to the left in the left eye and slightly off to the right in the right eye. I have Unity 5.5.0b10 and I tried many different combinations, (With or without the steamVR plugin, and many other configurations).

Typically if I open my project, set-up Steam VR prefab character, attach the post effects to the eye camera, I get this rendering issue.

I read on github that SSRR had issues in VR but I'm not sure this was the actual problem that it was talking about.

FYI other than that the scene is fluid and there are no particular issues.

 

Is there any way to preserve the depth buffer when postprocessing?

For VR, I want to render some in-scene UI without postprocessing, and then some controller models which are usually on top of that. I'm trying to do this by rendering a second camera without clearing Z for the UI - but the Z-buffer doesn't survive if any postprocessing is enabled.

I think this is a general OnRenderImage-related issue that's not specific to the new postprocessing system, but if there's a good solution to this, it'd be nice if it was built into the system?

 

I support this idea, it would be awesome to have this effect in the stack, and it would be even more awesome if it was a Unity's own version.

 

Has anyone else noticed a memory or performance leak while using this effects stack? When I first run a simple scene with this stack enabled everything runs very smooth at 70+ fps . However within a few minutes of running the scene, performance starts dropping into the 50s then 40's and even lower. This is without even doing much of anything in the scene apart from looking around. At first I thought it was specific to the SSR effect, however it does seem to happen with other effects as well, though not quite as badly or as quickly.

 

Is the limit of 300 units for the max distance in Screen Space Reflection going to stay this low? Or did I oversize my scenes? Because 300 units seem way to low for a top-down like camera perspective.

 

Thanks Chman I'll stick to Candela then until Unity solution is out (but Candela SSR is much much slower...)

 

laptop?

....heat?

 

Is dithering planned to be added? It was really great in the Cinematic effects and here it seems like an odd omission.

 

I've had this happen to me sadly

 

Does anybody work with Voxels (Voxel model to static mesh export)? What scale do you use for each voxel?

Mine is the default of the Voxel Importer extension, which is one unit per voxel. I didn't have distance problems in UE4 with this scale. But given that 300 units are max, and will stay this way. I'm going to have to shrink by /10.

The thing is, that both voxel importers on the store don't extend Unity's file import pipeline really. They just act as components or via separated editor panel generators. This also means that since Unity is not supporting nested prefabs, it's a hassle (even in code) trying to construct voxel models inside unity for those who have to, because they use MagicaVoxel that can't do constructions since it only got one single voxel grid.

 

@Chman First off, thanks so much for the amazing work, this new approach is light years better than the old, and the performance is really impressive.

I'm wondering if you're planning on adding multi-edit support?

Currently we have an effects stack that we want to use for the main VR camera, but also for the Spectator camera.

For the most part, the stacks are the same, but the Spectator stack has a few extra effects, specifically Depth Of Field and Vignette.

Currently, I'm using a shared profile, and then on my VR camera I Instantiate(profile) and then disable DoF and Vignette. It works fine, but prevents me from tweaking both stacks at runtime, which is a bummer.

Ideally I would just use 2 different profiles, but because they don't support multi-instance editing, this would be too cumbersome, as I'm constantly copy and pasting settings from one profile to the other trying to keep them in sync. If we could select both profiles, and tweak the shared settings we're interested in, that would be a nicer workflow.

 

Also, I'm curious about a couple of things, DoF seems to darken the image much more than in Blurs it, is that indended?
https://gyazo.com/64d9b04edec4d0fcfb8f139d29416348

And AO seems totally broken? This is all that ever shows in my game display:
https://gyazo.com/72f8f76a15c2fb8ae3adf5cb598946a0