I know this video is pretty old by now, but I still admire the nice sampling pattern in the Sponza scene (I'm trying to replicate it myself with Brigade 2). Could you tell me if it is using the Sobol/Van der Corput sampling sequence?
@SuperGastrocnemius It did not use QMC, just a random generator. A difference with Brigade 2 might be that I scrambled the path regeneration pixels instead of always restarting a sample at its initial pixel. This gives more uniform screen coverage.
@rouncer81 I already fixed the startup brightness problem by improving the progress estimator, making it less vulnerable to startup bias. See my other video on "GPU ERPT Refraction".
To be honest I'm not impressed. Even with noise the path traced architectural scene looks better. Pristine lighting quality. ERPT produces something that's not distinctly better than what a modern graphics engine would produce.
@4E65676174726F6E My theory is that you don't posses a sufficient amount of testosterone to be able to appreciate this. This is some very manly stuff, and some who lack this level of manliness might just not get it.
@4E65676174726F6E Modern graphics engines do this with precomputation. No graphics engines used in today's games would be able to produce the same image dynamically in realtime. This has the advantage of still producing physically correct images with dynamic lighting and objects, not static lighting maps that can only map lighting onto the non-changing environment.
I was strictly comparing to the path traced equivalent, just commenting on the relative visual quality of the resulting image.
The new videos however change the situation entirely. Dynamic range no longer appears to be smudged as it is here and it also efficiently renders occluded light sources and caustics. Indeed it appears to be superior to any GPU render I've ever seen.
@Dietger86 Real-time path tracing at HD resolution, ... well I'm stunned and speechless, never thought it would be possible this soon. Also, ERPT looks very promising, it's the first time that I see this algorithm in near real-time.
I know this video is pretty old by now, but I still admire the nice sampling pattern in the Sponza scene (I'm trying to replicate it myself with Brigade 2). Could you tell me if it is using the Sobol/Van der Corput sampling sequence?
SuperGastrocnemius 1 month ago
@SuperGastrocnemius It did not use QMC, just a random generator. A difference with Brigade 2 might be that I scrambled the path regeneration pixels instead of always restarting a sample at its initial pixel. This gives more uniform screen coverage.
Dietger86 1 month ago
@Dietger86 Thanks a lot. Will give it a try.
SuperGastrocnemius 1 month ago
god damn.. let it render a scene ffs..
MagnesiumAlloy 6 months ago
maybe you should make it go through a brightening filter at the beginning? i wonder what it would look like...
rouncer81 1 year ago
@rouncer81 I already fixed the startup brightness problem by improving the progress estimator, making it less vulnerable to startup bias. See my other video on "GPU ERPT Refraction".
Dietger86 1 year ago
To be honest I'm not impressed. Even with noise the path traced architectural scene looks better. Pristine lighting quality. ERPT produces something that's not distinctly better than what a modern graphics engine would produce.
4E65676174726F6E 1 year ago
@4E65676174726F6E My theory is that you don't posses a sufficient amount of testosterone to be able to appreciate this. This is some very manly stuff, and some who lack this level of manliness might just not get it.
Bartkei 1 year ago
@4E65676174726F6E Modern graphics engines do this with precomputation. No graphics engines used in today's games would be able to produce the same image dynamically in realtime. This has the advantage of still producing physically correct images with dynamic lighting and objects, not static lighting maps that can only map lighting onto the non-changing environment.
Phygar1 1 year ago
@Phygar1
I was strictly comparing to the path traced equivalent, just commenting on the relative visual quality of the resulting image.
The new videos however change the situation entirely. Dynamic range no longer appears to be smudged as it is here and it also efficiently renders occluded light sources and caustics. Indeed it appears to be superior to any GPU render I've ever seen.
4E65676174726F6E 1 year ago
Excellent job! It looks great! What's the render resolution in this video?
SuperGastrocnemius 1 year ago
@SuperGastrocnemius Thanks. The rendering resolution is 1280x720.
Dietger86 1 year ago
@Dietger86 Real-time path tracing at HD resolution, ... well I'm stunned and speechless, never thought it would be possible this soon. Also, ERPT looks very promising, it's the first time that I see this algorithm in near real-time.
SuperGastrocnemius 1 year ago
cool!
Bartkei 1 year ago