Env. Light Visibility Cache

[Dade]

I'm working on a new version of the "Visibility driven sampling of env. light sources (aka portals without portals)" (viewtopic.php?f=5&t=93).
The original version was using a single visibility map for all the scene. The new version instead uses many visibility maps, one for each cache entry with the entries evenly spread over all surfaces as usual (i.e. the same of PGIC and DLSC).

Indeed, being able to have a local visibility map, is far more accurate than using a global one. The draw back is an increase of memory required to store all maps.

Today, I have done the very first rendering with the new code. This is a direct light only rendering (100 samples/pixel) without visibility map:

and this with the new EVL cache:

It looks promising.

[Sharlybg]

Uhm ! juicy result for an early code ! congratulation Dade .

[lacilaci]

Looking good...

1. Shadows in the back wall still look quite noisy, why is that?

2. How bad is the memory footprint? Does it increase with scene complexity?

3. Can this be used for other light types too?

[Sharlybg]

1. Shadows in the back wall still look quite noisy, why is that?

Maybe not enough sample. Was not intended to be 100%
Clear as it is a direct comparison at same sample level.

[lacilaci]

1. Shadows in the back wall still look quite noisy, why is that?

Maybe not enough sample. Was npt intended to be 100%
Clear as it is a direct comparison at same sampne level.

Maybe, I'm also viewing it on a phone atm...

Anyway, if there's a build I'll give it a proper try tomorrow, cause today I saw only an unsuccessful build on azure.

[epilectrolytics]

I'm working on a new version of the "Visibility driven sampling of env. light sources (aka portals without portals)"

Indeed it looks promising!

The data collected by tracing photons are a gold mine: at the moment I'm in doubt between a new version of visibility map for infinite lights or guided brute force path tracing (when current stuff is done).

Is "guided brute force path tracing" still considered for development later?

[Dade]

2. How bad is the memory footprint? Does it increase with scene complexity?

It was my main concern at the start however I haven't seen a noticeable increase of memory usage in this scene (it has about 1300 cache entries).

The complexity of the scene will increase the amount of cache entries required however it seems manageable at the moment.

3. Can this be used for other light types too?

It could in the future (see my next post).

[Dade]

The data collected by tracing photons are a gold mine: at the moment I'm in doubt between a new version of visibility map for infinite lights or guided brute force path tracing (when current stuff is done).

Is "guided brute force path tracing" still considered for development later?

This method is very similar to a guided method. It "guides" only for env. light sources but it could "guides" for any kind of incoming light (direct and indirect).

[Piita]

Looks promising indeed

[lacilaci]

Is this enabled by default?(replacing current visibility map) or is it a new feature/mode that has to be exposed in blender?