Direct Light cache: how to avoid splotches?

[Dade]

Isn't this always the same bug? I mean visually it always looks like the MIS issue with caching, or is it something different?

Only if it is glossy and I asked before (posting.php?mode=quote&f=4&p=8053#pr7607) to redo the test with mate material so I assume the above rendering is showing a different problem

[lacilaci]

Isn't this always the same bug? I mean visually it always looks like the MIS issue with caching, or is it something different?

Only if it is glossy and I asked before (posting.php?mode=quote&f=4&p=8053#pr7607) to redo the test with mate material so I assume the above rendering is showing a different problem

Well there's also the other issue which looks a lot like this one, when you use heavy clamping.. Remember? I think it's always the same thing going on, just very visible with heavy clamping and/or very rough reflections + multiple light sources.

[lacilaci]

And what's up with the spot made of bright pixels here? Related?

[marcatore]

Isn't this always the same bug? I mean visually it always looks like the MIS issue with caching, or is it something different?

Only if it is glossy and I asked before (posting.php?mode=quote&f=4&p=8053#pr7607) to redo the test with mate material so I assume the above rendering is showing a different problem

I sent you a PM with the scene.
About the carpet on the floor I confirm I used a matte material

[Dade]

And what's up with the spot made of bright pixels here? Related?

The first step when checking a DLSC related problem is always to switch max. path depth to 1 so you will see only direct light sampling:

It becomes quite evident that at glancing angles the information of the cache are not correct due to "normal threshold" (10 degrees in this case). If you lower the normal angle threshold to 4 degrees (and increase the radius to 15cm just to speed up the pre-processing):

and with full path depth:

[Dade]

I sent you a PM with the scene.
About the carpet on the floor I confirm I used a matte material

Thanks, going to check but I have the strong feeling the problem is in the impossibility to have reliable cache information over the carpet (the particle system for hair simulating the carpet will likely to generate a surface over which is impossible to interpolate correct information).

[marcatore]

Some exclude list or something else to help the engine with samples estimation?

Could be helpful in this case?

[lacilaci]

Some exclude list or something else to help the engine with samples estimation?

Could be helpful in this case?

Or maybe retracing problematic areas if they can be identified...?

[FarbigeWelt]

And what's up with the spot made of bright pixels here? Related?

The first step when checking a DLSC related problem is always to switch max. path depth to 1 so you will see only direct light sampling:

prob1.jpg

It becomes quite evident that at glancing angles the information of the cache are not correct due to "normal threshold" (10 degrees in this case). If you lower the normal angle threshold to 4 degrees (and increase the radius to 15cm just to speed up the pre-processing):

fix1.jpg

and with full path depth:

fix2.jpg

Wow! This is a good step by step instruction for DLSC related issues. Thank you much, Dade.

[lacilaci]

If I remember correctly, vray has this retracing threshold value for Light caching, I guess it's a workaround to deal with these type of problematic areas...? If we could detect shadow borders and these areas where cache has problems, maybe we could force "retracing" those areas, replacig cache with brute force?