But boy howdy do they get results!
What model are real-time people favoring for skydome radiance these days? Hosek-Wilkie? /cc @meshula
-
-
-
Just did an unironic search for http://TrackTrack.It .
-
It’s dead, Jim. Let it go. /cc
@philmccarty -
But hey, I get to compute that myself?
End of conversation
New conversation -
-
-
Cheap’n’fake for me. (Will probably work out something cheap and decent at some point, but right now I’m low on GPU cycles.)
-
Check out - https://github.com/ebruneton/clear-sky-models … to compare models, and https://github.com/ebruneton/precomputed_atmospheric_scattering … for
@EricBruneton's model. I'm still liking Hosek-Wilkie, and I'm enjoying Eric's model at the moment. -
Oh, this is a nice collection, thanks! The few I’ve taken a quick look at look far off my needs (evaluation many times per pixel, GPU already nearly on fire :) but could perhaps be approximated.
-
I hear that :) Generally skies change kinda slowly. A strategy I used is to write the values into a 32x32 cubemap on the CPU in a low priority background thread and upload the cube to GPU whenever a new cube is ready. (16x16 visibly bands, 64x64 is only a bit nicer)
-
Yeah, this sounds about right. Definitely not trying to evaluate or multiply-sample the model every frame. Thanks for the pointers!
-
Hmm… good points. I guess this can be combined with a simple disk for when the sun is visible (don’t want it pixellated :) Then it’s just a few hundred cube map samples per pixel. Still ouch, but much reduced.
-
You really need the disc, most (all?) of the models generate the skyglow around the sun, but not the sun itself. Can you reduce the sampling load by convolving some of your calculations into a mip-chain? I don't really know what you're doing so I'm spitballing :)
-
Realtime path tracing, so it’s a case of sampling a lot of rays per pixel, and each ray will also sample the sky multiple times to determine lighting. So Kind of unavoidable :)
End of conversation
New conversation -
-
-
0:16I BLEW THE ROOF OFF! Finally... Rayleigh and Mie scattering (O'Neil 2004, First GPU GEMS)
Cascaded-SparseVoxel hybrid #pathtracing w/ naive spatial de-noising... (30-94 FPS, nV 970M GTX,1024x768). More details in reply.#3d#rendering#vulkan#vulkanapi#gamedev pic.twitter.com/ioc9g7RnjzShow this threadThanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.