Page 1 of 1

[News] PCC / VCT hybrid progress

Posted: Wed Aug 14, 2019 6:02 pm
by dark_sylinc

Re: [News] PCC / VCT hybrid progress

Posted: Wed Aug 14, 2019 11:02 pm
by al2950
Interesting read :)

One question. I am curious how SSR might or might not fit into all this..?

Re: [News] PCC / VCT hybrid progress

Posted: Thu Aug 15, 2019 12:31 am
by dark_sylinc
In my experience SSR was fun to implement and a terrible disappointment.

It looks really good on still images, but it fails terribly in motion.

At first I thought our implementation was bad (that could be half true), but then I saw popular games using SSR and I saw horribly distracting SSR artifacts, the same we were having.

State of the art stochastic temporal SSR as implemented by EA looked like much superior, as shown by the papers. I was reluctant to implement it due to its complexity but seemed like it was worth the trouble. Then I tested a few EA games where it is supposedly used and it's.... not good. Too noisy for real time objects, and many of the artifacts are just hidden with noise, to the point it feels like you're watching a poorly tuned TV station.

Thus yes, SSR could be used and somehow be part of a hybrid, but it doesn't seem to be worth the trouble when accounting the artifacts, stress on the artists, performance impact, and overhead of constantly improving the code.

SSR can't get over the big problem that if you're looking perpendicular to a wall, then SSR shines and captures almost everything in the reflection; until you turn 90° to look straight into the wall and not a single pixel in the wall gets benefit from SSR. The algorithm is extremely unstable to camera motion.

And then there's the problem that SSR is terrible for 3rd person games. The player's character/car/plane/ship blocks vital information required by SSR, causing horrible artifacts and these are present right where the user will be focusing his eyes.

Quite possibly the only niche where SSR is useful is for distant objects, which often get into the ideal case for SSR, are too small to notice any artifacts, and get to cover the areas where there is no voxel or cubemap information.

Re: [News] PCC / VCT hybrid progress

Posted: Mon Aug 19, 2019 5:19 pm
by TaaTT4
Despite the defects highlighted by Matias, I believe that SSR will be the right choice in my game. First of all, my levels consist in outdoor only scenes with a lot of distant objects (it's a racing game!). And I doubt that I can benfit of GI in a scenario like this. Secondarily, I have poor artist control over levels since they are procedurally generated. I reasoned about placing some PCCs along the track and in its surroundings, but I fear this approach. I guess I would need a lot of PCCs to cover all the areas where the car can go an this would kill video memory.