Cache-less solutions for animations

General project and community related discussions and offtopic threads.
Post Reply
User avatar
lacilaci
Donor
Posts: 1928
Joined: Fri May 04, 2018 5:16 am

Cache-less solutions for animations

Post by lacilaci » Sat Apr 04, 2020 7:41 am

So while caches are very useful for still images and also complex caustics simulations. They are very unwieldy for animations and not usable at all in some situations.

For this reason i had to render several quick small animations outside luxcore since instability, lack of speed and the need for careful setup was making luxcore a "problematic" choice.

So, there are few things/options for quick animations and fast preview rendering that could help a lot to make luxcore attractive to people that do a lot of anims and iterations of design.

For these purposes there are 3 main problems.

1. Direct light noise:

I recently posted here on forums a case, where env. Cache failed even on still image and produced artifacts that wouldn't go away. But aside that, this cache is not possible to use in animations where objects and/or lights are moving.

For animations and some problematic cases, portals would be much better solution.

2. GI noise:

While to superior to any other solution for GI noise, it is simply not possible to use it in animations where objects and/or lights are moving. I don't know if there is a stable and also easy to use alternative, but maybe cuda and rtx support (if they provide substantial boost) could bring luxcore performance on a comparable level to octane or e-cycles.

3. Fireflies:

Sharly posted recently an interesting comparison on how octane clamping handled not only fireflies but also overall noise very efficiently without drastically affecting overall lighting. Improved clamping could not only provide better way for luxcore to handle fireflies but also help with problem #2 a bit.

I dont think caches could be improved to a point where complex animation is a: easy to setup, b: stable enough so user has no worries that some frames will contain problems. Especially if we consider that sometimes they cannot help even in a still frame case without introducing unsolvable artifacting(env cache) and require carefull preparation of scene(photongi).

User avatar
Sharlybg
Donor
Posts: 2082
Joined: Mon Dec 04, 2017 10:11 pm
Location: Ivory Coast

Re: Cache-less solutions for animations

Post by Sharlybg » Sat Apr 04, 2020 9:03 am

For me the solution to this was obvious since we start to develop Cached solution. Always talk about it. Rendering with progressive importance sampling Aka Online learning path guided sampling. With that we can wait easily for all the other trick to be develop.

1/ It is bloodly Fast something between 8X to 16X faster

2/ It work Even in extreme case where light is hardly reachable

3/ Work without artefact

4/ It is viable for CPU based solution (PathCPU __BidirCPU)

5/ Can work side by side with PGI on CPU for Still or Camera animated Project (Hyper light speed on :twisted: )

6/ A good alternative to PGI for Light Groups

7/ There is a good paper for that

Image

Image

https://corona-renderer.com/blog/resear ... -sampling/
Support LuxCoreRender project with salts and bounties

Portfolio : https://www.behance.net/DRAVIA

epilectrolytics
Donor
Posts: 703
Joined: Thu Oct 04, 2018 6:06 am

Re: Cache-less solutions for animations

Post by epilectrolytics » Sat Apr 25, 2020 4:51 pm

lacilaci wrote:
Sat Apr 04, 2020 7:41 am
2. GI noise:

While to superior to any other solution for GI noise, it is simply not possible to use it in animations where objects and/or lights are moving. I don't know if there is a stable and also easy to use alternative, but maybe cuda and rtx support (if they provide substantial boost) could bring luxcore performance on a comparable level to octane or e-cycles.

I dont think caches could be improved to a point where complex animation is a: easy to setup, b: stable enough so user has no worries that some frames will contain problems. Especially if we consider that sometimes they cannot help even in a still frame case without introducing unsolvable artifacting(env cache) and require carefull preparation of scene(photongi).
Perhaps a periodic indirect PGI cache similar to what the caustic cache already does would help (but more a bit than a lot).
Several indirect caches with different seeds would reduce bias and make animation of objects and lights possible.
Also the glossy (ghost) reflection problem would be addressed.
Periodic PGI could also made faster by calculating the cache iterations parallel to GPU rendering, like light tracing, instead of alternating which is the way it works now.

It is no solution for all problems but an improvement I'd like to test.

There is a firefly problem from the metropolis light-tracing that produces the cache, it shows in all renders using SDS-cache I do.
Perhaps a separate clamping setting for PGI could help with that.

Post Reply