8

Recently while searching of solving the shadow acne problem of traditional shadow mapping I implemented exponential shadow maps. It did the trick (no acne at all) but at the same time introduced other unacceptable errors.

For depth map blur I use gaussian blur with smallest sigma=1.

Test for shadow map:

float occluder = texture(shadowMap lightCoords.xy).r;
float c = 5000.0;
float receiver = lightCoords.z;
float shadow = exp(c*(occluder-receiver));
shadow = clamp(shadow, 0.0, 1.0);

with small c factor c=100.0:

enter image description here

Unacceptable light bleeding

with high c factor c=5000.0:

enter image description here

No light bleeding but high frequency detail appear "swollen".

Cant find optimal c - light bleeding appear even at c=3000.0 and high frequency shadow error is already there. Not blurring the depth map helps but then aliasing occurs.

And my question is - how can I improve this technique (ESM) or should I look for another? I think good quality shadows with nice performance are definitely possible - as seen in modern games like Witcher 3 or Fallout 4.

Community
  • 1
mdkdy
  • 2,169
  • 1
  • 12
  • 19

1 Answers1

2

Well nevermind. I made some research and actually gonna:

  • sample ESM using Poisson sampling in PCF fashion (thanks Bart Wronski for great Poisson Sample generator app) and extend this to PCSS technique (variable penumbra). Render to texture.
  • optionally depth aware blur to soften - EDIT not so fast this option doesnt work as expected, too bad. Instead rotating PCF kernel seem to work fine.

And too bad not much of discussion here.

mdkdy
  • 2,169
  • 1
  • 12
  • 19