Dynamic shadows artifacts

[jonimake]

Would slope bias help with cases like this? http://i.imgur.com/SLTGKHs.png

[Zeblote]

Would slope bias help with cases like this? http://i.imgur.com/SLTGKHs.png

Yes. Fixing things like that is what slope bias is for.

[jonimake]

Yes. Fixing things like that is what slope bias is for.

Well then I'm quite astonished, since everything suffers from it. Directional lights look particularly ugly on relatively clean flat surfaces when the surface and light direction are close to perpendicular. http://i.imgur.com/pXwuHX2.png

[Maximum-Dev]

[MENTION=738]Daniel.Wenograd[/MENTION], Yes I remember we had to manually download resource compiler form the thread and copy it to it's location.

Regarding DFAO, CryEngine 2 (2007) had sky occlusion, very cheap and clean, without any artifacts regardless of mesh scale. http://docs.cryengine.com/display/SD...ient+Occlusion
And now CryEngine 5 has large scale AO + SVOGI. http://docs.cryengine.com/download/a...5668000&api=v2

What I don't understand is, Why Cryengine with 16 landscape layers, POM everywhere with 256 steps, Dynamic lights, Volumetric Lighting, Volumetric fog, Volumetric Clouds, Dynamic shadows with 7 cascades, Real time G.I, Large scale AO, SSDO and all that stuff, still runs -faster- than UE4.

I'm not pro crytek or pro cryengine, in fact I have very opposite thoughts with crytek. But when they do something with engine, they do it properly. But here when you say why this is slow, most answers are like Ooooch! you have 10 dynamic lights you should cut them down or turn off shadow casting! But nobody thinks about the fact that 14,716 dynamic lights were placed throughout Crysis 3's 8 levels, 54.55% of them were shadow casting (think about the numbers), meaning out of 14,716 dynamic lights around 8027 of them were shadow casting lights, divided by 8 levels it means 1 map had around -1000- shadow casting lights and -827- non shadow casting lights. In the same game, in an interior, -150- dynamic lights cost only 4ms and that's on an outdated GTX 680 that's way below the bottom of the barrel today. Just think about the numbers.

With that said, I have no idea why Epic does not try to improve dynamic lighting or even don't want to acknowledge there's anything wrong with it, let alone the improvement. At least can someone from Epic leave a reply on this 4 year old shadowing problem.

[Zeblote]

Well then I'm quite astonished, since everything suffers from it. Directional lights look particularly ugly on relatively clean flat surfaces when the surface and light direction are close to perpendicular. http://i.imgur.com/pXwuHX2.png

I was just as surprised as you are now that unreal doesn't have this standard feature. It makes no sense at all.

[jonimake]

I'd appreciate if someone from Epic would at least comment on this.

[Maximum-Dev]

[MENTION=16769]jonimake[/MENTION]

[Deathrey]

Difference between using and not using slope bias is kinda appreciable.


Reference shot with a typical depth bias setting:



Removing acne from unshadowed surfaces with slope bias without increasing depth bias:



Removing acne from unshadowed surfaces using depth bias only for comparison:



Notice how the small sphere starts peter-panning and its shadow becomes smaller than it should be on the last shot, and compare it with a slight distortion near contact point on a second shot.
This is pretty much brief gist of all the thread up to this point, and as a conclusion I'd add that being able to adjust both slope bias and depth bias separately for each cascade is desirable.

[Maximum-Dev]

That stepped shadows we get on flat surfaces is worse than the blocky shadows on round surfaces IMO.

[Zeblote]

How did you take the screenshots? Since noone from epic seems to give a ****, maybe you could upload a pull request with this?

[Deathrey]

How did you take the screenshots?

These shots were not from UE.
Here is corresponding implementation and roughly the same test case for UE4(scale of objects might be a bit different):

Reference shot in UE4 with default depth bias setting (0.5) :

 

Spoiler

UE4 Depth bias(0.5), Slope Bias (10.0) :

 

Spoiler

UE4 Depth bias(4.0), Slope Bias (0.0) :

 

Spoiler

In shot 2 and 3 Slope Bias and Depth Bias respectively were increased until artifacts on background wall were not distinguishable anymore.
Once again, the shadow shape change is quite noticeable with same visual results between shots 2 and 3.

maybe you could upload a pull request with this?

What I have so far can be safely labeled as shitcode and is definitely below the standards for a proper pull request. I'd be willing to bring it to a decent state at some point though, provided I have enough free time.

Here is an answerhub feature request in the meantime.

[jonimake]

Maybe no Epic dev looks here because the title says 4.14 even though this affects all versions. (sneaky bump)

[StrangerGwenn]

In Helium Rain, this problem was so bad for planets that we ended up with unlit shading and some math to generate lighting. I would so love for this to get in the engine.

[Maximum-Dev]

Maybe no Epic dev looks here because the title says 4.14 even though this affects all versions. (sneaky bump)

They should be curious why this is keep bumped though right?

[Deathrey]

[MENTION=30269]netfrag-sam[/MENTION] Would it be much of a hassle for You to change the thread name to something like "Whole scene dynamic shadows artifacts" ?
4.14 in the title might be somewhat misleading, because the problem affects all versions.