Struggling to Convert Global Position to Shadow Map position

Pryme8 · November 11, 2019, 10:38pm

I am trying to get a global position to the shadow map uv position for a ShadowGenerator Texture.

float sampleSunShadow(vec3 pos){		
		vec4 posFromSun = vec4(pos, distance(sunPosition, pos));		
		vec3 clipSpace = posFromSun.xyz / posFromSun.w;		
		vec2 uv = 0.5 * clipSpace.xy + vec2(0.5);
        if (uv.x < 0. || uv.x > 1.0 || uv.y < 0. || uv.y > 1.0)
        {
            return 1.0;
        }		
	    return texture2D(sunShadowMap, uv).x;
	}

This is what I am currently doing, but it seems to not really work, I have also tried quite a few other methods… figure it just getting the correct combo of matrices.

Deltakosh · November 11, 2019, 10:45pm

Did you check the shader we use for PBR or stdMat?

Pryme8 · November 11, 2019, 10:48pm

yeah, and when I try to duplicate it I seem to still have a sample issue.
Maybe it’s the global position that I am messing up.

Ill check if the PBR example helps me.

hmm its the same function, and as far as I can tell I am using the same process… dang gotta figure this out.

Deltakosh · November 11, 2019, 11:00pm

Maybe the position or the light matrix?

Pryme8 · November 11, 2019, 11:06pm

Im super close, let me see if I can post a PG for it. Somebody might see what’s good right away.

https://playground.babylonjs.com/#21FNQE#10

its on line 565 where I am sampling the shadowmap, I have a sneaking suspicion I am just sampling the wrong global point or the incorrect perspective amount.

Deltakosh · November 12, 2019, 12:04am

why this:

length(pos+sunPosition)+50./600.)

Pryme8 · November 12, 2019, 1:00am

I am assuming that is the way to convert the depth of the rayPosition to the Light, but might be wrong.

My shadowMin and max is 50 and 600 units

Ive tried

vec4(pos, distance(pos,sunPosition));

as well.

Evgeni_Popov · November 12, 2019, 10:06am

Just to be sure I understand what you want to do:

You want to shadow the clouds that are hidden (blocked by geometries) from the point of view of the sun?

If yes, what I don’t understand is that you take the value from the shadow map (return texture2D(sunShadowMap, uv).x;) as the light attenuation in your lighting equation.

However, the value in the shadow map is a depth from the light point of view (the depth of the geometry nearest to the light). So, in your code, it means the farther the point is from the light, the darker it is, being in shadow or not (meaning no geometry blocking the path of light).

What you need to do is to transform your point into the light (sun) coordinate system and get the depth of this point in this system (the z value). If the depth is less than the one in the shadow map for this location, then the point is not in shadow (so the factor in your lighting equation is 1.), else it is (use whatever factor you want to darken the point).

Going to give a try to do this, but don’t hold your breath…

Pryme8 · November 12, 2019, 1:37pm

if we can figure this out then I will optimize the whole system to run at 60 fps, and then introduce simulation shaders to make it so we can have some real volumetric smoke and other things.

I think I know what you are saying, I was mistaken about what the shadowmaps data had on it! I did not realise it was a depth value, now that I know that though we might be able to make this happen.

Pryme8 · November 12, 2019, 2:23pm

wow I was doing a bunch wrong now that I am looking at it with fresh eyes.

Pryme8 · November 12, 2019, 2:52pm

This is the closest I have gotten now.

https://playground.babylonjs.com/#21FNQE#15

And that is still off… I have tried using the lights matrix, the shadowGenerators Matrix hmmm there is gonna be an answer and once its figured out this is gonna look cool…

I think it might be I am sampling the wrong position still, which I am troubleshooting now as I drink me coffeeeee…

Evgeni_Popov · November 12, 2019, 3:44pm

I think DepthMetric should not be 1.0 in the sampleSunShadow function.

It should be the depth of your point (pos) in the light coordinate system (LCS), meaning the z value after pos is transformed into the LCS. For this you need the light matrix.

Also I don’t really understand the calculation with the sun and point positions, being normalized (?)

Pryme8 · November 12, 2019, 3:46pm

I don’t either, I’m just guessing at this point… I’ve tried like every combination I can think of.

I’m thinking my sample position was off so I am going back and checking that I am feeling the correct global position in then I am going to try to multiply it by the shadowGenerators TransformMatrix agian and see if I am just being dumb.

Im convinced that this:

vec4 p = shadowMapTransformMatrix * vec4(pos+boxOffset, 1.0);
float depthMetric = 1.0;
vec3 clipSpace = p.xyz / p.w;
vec2 uv = 0.5 * clipSpace.xy + vec2(0.5);

Should work… but maybe I am just using the wrong matrix?

Pryme8 · November 12, 2019, 4:10pm

Ok so the output position is correct.

...
if (density > 0.0) {			
return vec4(rayPos+boxOffset, 1.0);
...

If I apply a matrix like view I get the expected results.

...
if (density > 0.0) {			
return view*vec4(rayPos+boxOffset, 1.0);
...

Then if I apply the shadowGens Transform Matrix all the “clouds” disapear…

...
if (density > 0.0) {			
return shadowMapTransformMatrix*vec4(rayPos+boxOffset, 1.0);
...

Maybe thats correct? Not sure… I am about to contimplate a different maching method if I have time here before I gotta get on other things.

Evgeni_Popov · November 12, 2019, 4:42pm

Here it is: https://playground.babylonjs.com/#21FNQE#16

Note that shadowMapTransformMatrix was not in the uniform list passed to BABYLON.PostProcess, so you ended up with an all-0 matrix in the shader.

I mostly copied what is done in the computeShadow method used by the shadow fragment shader used by Babylon.

Pryme8 · November 12, 2019, 4:44pm

GOD I LOVE YOU! This is amazing, now to optimize.

I knew it was something stupid…

Evgeni_Popov · November 12, 2019, 4:58pm

Calling sampleSunShadow is killing perf: I’m at 60fps in fullscreen without it, 20fps else…

Pryme8 · November 12, 2019, 7:04pm

Fixing all that now!
I’ve got it performing fairly well now considering its procedural noise.

aFalcon · November 12, 2019, 9:42pm

@Pryme8 - RayMarcher code is such a GREAT read.

> vec2 cloudMarchSunLight(vec2 rayData, float depth, vec3 offsetDir){

So elegant.

And references: nice!

LIKE the header references: GOOD-PRACTICE, good~links:

// Cloud FX
// Pryme8
// refrencing https://www.youtube.com/watch?v=4QOcCGI6xOU
// and
// Post Page

//WIll do that too.

Hello PERLIN Noise Shadertoy. : )

And the FIX??? Was?

REDCORNER: POSTPROCESSES!

AND BLUECORNER: uniform mat4 shadowMapTransformMatrix!

And that’s what I got from that.

This is getting good…

Can’t wait for next round.

: )

Pryme8 · November 12, 2019, 9:48pm

It ended up being that I was not passing the Matrix to the shader >_< missed one tiny line of code… hahaha

Topic		Replies	Views
"Clipping cuboid" to remove rendererd output from an area? Questions	3	515	May 28, 2020
Particle Vertex Shader / Screen or View Position Questions particles , shader	17	1487	May 23, 2020
How to make the worldPos sampler texture and offset in the fragment shader Questions material , shader	3	477	April 23, 2023
ShadowGenerator: SIZE, Position, AND multiple instances? Questions	13	1622	February 13, 2020
Multiple Sun Shadows, is this right? Questions	32	1494	February 13, 2020

Struggling to Convert Global Position to Shadow Map position

Related topics