KD tree bug - again

Post by **fused** » Sun Mar 29, 2009 9:07 am

with version 1.1.18 x64.

BVH is ok, kd tree misses lots of intersections

(also, there are quite a few fireflies and the longer i render the more fireflies turn up. weird, because i only used diffuse materials. pathtracing without bidir)

Kram1032 · Post by **Kram1032** » Sun Mar 29, 2009 10:41 pm

at least, they don't turn black anymore... xD

PureSpider · Post by **PureSpider** » Tue Mar 31, 2009 4:35 am

They just turned black cause not much light got inside the poly-holes... Same issue as far as I see it

Post by **fused** » Tue Mar 31, 2009 8:52 am

Did some testing on the firefly issue.
All renders are with linear tonemapping.

---------------------------------------------------------------------------------------
Blackbody emitters, PT.

---------------------------------------------------------------------------------------
Blackbody emitters, bidir PT.

---------------------------------------------------------------------------------------
Blackbody emitters, MLT.

---------------------------------------------------------------------------------------
Blackbody emitters, bidir MLT. note how different it looks compared to the other blackbody versions.

---------------------------------------------------------------------------------------
Sun, MLT.

---------------------------------------------------------------------------------------
Sun, PT.

---------------------------------------------------------------------------------------
RGB emitters, PT.

---------------------------------------------------------------------------------------
RGB emitters, MLT.
Emitters are scalted by the factor 5 this time. Fireflies disappear.

---------------------------------------------------------------------------------------
RGB emitters, PT.
Emitters are scalted by the factor 5 this time. Fireflies disappear.

Kram1032 · Post by **Kram1032** » Wed Apr 01, 2009 1:44 am

very nice test

So, the kd-problem is solved? Or are they all with bvh?

Post by **fused** » Wed Apr 01, 2009 3:07 am

all with bhv, kd is still borked.

but ono said that hell fix the firefly issue.

SATtva_ · Post by **SATtva_** » Wed Apr 01, 2009 3:14 am

I hope you meant "he'll". :)

Post by **fused** » Wed Apr 01, 2009 3:17 am

CTZn · Post by **CTZn** » Wed Apr 01, 2009 3:17 am

Everyone wants chaos to be under control, that's insane

Damnit, PT is PT... or is it really "broken" ? Though, any fix is a good thing.

Post by **fused** » Wed Apr 01, 2009 3:18 am

CTZn wrote:Damnit, PT is PT... or is it really "broken" ? Though, any fix is a good thing.

well, obviously its not working correctly when using small emitters. "broken" is probably different

CTZn · Post by **CTZn** » Wed Apr 01, 2009 3:29 am

I don't see this as being incorrect, just unwanted

It would need an expert to confirm or infirm my says, but I believe that simple PT is archaic and that fireflies are the normal result of that stochastic(=random?) process wich uses no guides (optimisations) afaik. So reducing or removing fireflies would change the nature of pt to something different, that's my belief. I'm wondering if the request is sane, really.

When I want to reduce path tracing limitations to a maximum, I use plain MLT

Well bidir does a great job too isn't it ?

Kram1032 · Post by **Kram1032** » Wed Apr 01, 2009 4:28 am

tiny light-sources often make problems...
As it's rendering from Camera to Lightsource (to avoid dead rays), Indigo needs to randomly find a lightsource. A big emitting area allows faster convergance, as it's more likely to be found.
That's also, why Indigo with more lightsources rather gets more efficient than less. - the more direct light, the less indirect one

Post by **fused** » Wed Apr 01, 2009 4:54 am

Kram1032 wrote:tiny light-sources often make problems...
As it's rendering from Camera to Lightsource (to avoid dead rays), Indigo needs to randomly find a lightsource. A big emitting area allows faster convergance, as it's more likely to be found.
That's also, why Indigo with more lightsources rather gets more efficient than less. - the more direct light, the less indirect one

yep accepted and confirmed

CTZn wrote:I don't see this as being incorrect, just unwanted

It would need an expert to confirm or infirm my says, but I believe that simple PT is archaic and that fireflies are the normal result of that stochastic(=random?) process wich uses no guides (optimisations) afaik. So reducing or removing fireflies would change the nature of pt to something different, that's my belief. I'm wondering if the request is sane, really.

When I want to reduce path tracing limitations to a maximum, I use plain MLT Well bidir does a great job too isn't it ?

sorry ctz, but this doesnt sound logical for me. also, you know ono and how he is having and eye on staying as physically correct as possible. i guess if this was a usual limitation of PT he wouldn't have said that i can and will be fixed?

regarding plain MLT: take a closer look, it has fireflies, too!

CTZn · Post by **CTZn** » Wed Apr 01, 2009 9:05 am

Well, my logic beneath is that having fireflies in a render does not mean it's biased. If you have 3 fireflies on an image that means 3 unwanted samples amongst thousands, as you know. That is acceptable for me, furthermore because supersampling gets rid of ff like a charm.

Again I'm not an expert, I'm exposing own beliefs more than arguing and I would like to know more on that topic (are ff inherent to pt ?)

edit: on the MLT render with ff: is that reproducible ?

Post by **fused** » Wed Apr 01, 2009 9:21 am

download the scene and try it

the longer you render the more fireflies turn up and won't go away. this is definitely an issue

KD tree bug - again

KD tree bug - again

Who is online