Log in

No account? Create an account
I said I wasn't going to do anything more to the renderer right now.… - Spectre Gruedorf Challenge Blog [entries|archive|friends|userinfo]
Spectre Gruedorf Challenge Blog

[ userinfo | livejournal userinfo ]
[ archive | journal archive ]

[Mar. 22nd, 2009|07:24 pm]
Spectre Gruedorf Challenge Blog


I said I wasn't going to do anything more to the renderer right now. I lied. (We will stop enhancing the truth in 3...2... *BZZT*)

Dropped in variance shadow maps. Took me all of half an hour, and it improves shadow quality tremendously. What a clever idea. No blur pass yet, because I'm really a lazy sod, but even with just plain linear filtering you get good results.

I have to figure out why the hell my bias in the variance term is so fucking high, though, and I also have to figure out a way to try and clean up artifacts on the edges of 'sharp' pieces of geometry that are coplanar to the light. (i.e. the tops of crates and things) Nothing is perfect, but it's not bad...

[User Picture]From: nothings
2009-03-23 02:50 am (UTC)
Another thing to look at is "exponential shadow maps".

The whole thing is such a mess.
(Reply) (Thread)
[User Picture]From: mad_and_crazy
2009-03-23 02:59 am (UTC)
That's the same guy, isn't it? Andrew Lauritzen. Actually, he used to live in Victoria, oddly enough.
(Reply) (Parent) (Thread)
[User Picture]From: mad_and_crazy
2009-03-23 03:10 am (UTC)
Nope, I tell a fib. Exponential shadow maps are not one of Andrew's tricks.

I agree with you on the mess sentiment, though. I think getting point lights to shadow nicely is going to be a pain in the ass.
(Reply) (Parent) (Thread)
[User Picture]From: nothings
2009-03-23 03:21 am (UTC)
ESM was apparently developed independently by Salvi and by Annen/Metens/Seidel/Flerackers/Kautz. It supposedly doesn't suffer from the light leaking problem of VSM (which is frustrating because it's a real problem which appears to be really obvious if you use it in practice but was under-reported initially).

The paper with the five guys above points out that ESM does have a problematic case, but they can guesstimate when the problematic case occurs and then sample more finely to reduce the effect. Unfortunately the guestimate isn't perfect! Hence the mess.
(Reply) (Parent) (Thread)