Lightmass is the most correct one, even if it could have been baked at a higher quality.
DFGI is plain wrong, it’s reflecting green for whatever reason instead of orange, and the orange doesn’t radiate on other surfaces.
I’m always surprised about how much LPV looks good on simple scenes, it’s unfortunate that it leaks so much for generic environments.
I still want to see a real representative demo of DFGI working on a video, but unfortunately everytime i think about compiling unreal from scratch my nose bleeds
Yep DFGI is completely busted in that comparison, I’ll setup a comparative scene here and see what’s up when I get a chance. Lightmass looks good. LPV has color bleeding but lots of leaking as usual.
Well cascaded voxel cone tracing is working on ps4 in the tomorrow children, it really only works for there game(stylized, large objects to avoid leaking) but there reflections are really cool,(accelerated using distance fields ;))
I found something very interesting(to me ),I don’t know if this is any help but I found this http://www.kevinbeason.com/smallpt/
It is open source, all the code is there,and it is written in C++.
NO , that library is only for non realtime path tracing , trying to use it for games would be rediculous since is designed to be unbiased , not to be fast.
+1 - there is no other way. Every new honorable title just have it. GTA 5, The Witcher 3, Fallout 4, The Division. Hardware also getting there. Games just become bigger, more like sandboxes and backing is not the way it should be done with lighting. I’m not saying about effects - volumetric stuff, weather system, physics. Look at the ARK - its so heavy on resources and looking so 2010-ish.
Each of those games have a solution that really only works for that game. Until hardware gets better that will allow most people to use VXGI/SVOGI/etc. then the best you can hope for is a low-quality solution that will work with only a limited number of game types.
As far as I understood:
-no need for voxelization, uses the actual polygonal mesh
-not view-dependent
-gives Global Illumination, Ambient Occlusion and Caustics with one solution
It is from 2010 but I think with todays hardware, dx12 and optimizations (doing the path tracing part on the gpu rather on the cpu as the paper proposes) this technique could be even more performant.
I see that screenspace deep gbuffer method seems to work really well and fast, its only problem is being screen space. Would rendering a scene from one or two more views in a very low quality really impact performance that much? They wouldn’t be full quality passes, just very basic LOD meshes (only very close ones or far big ones being rendered) and low textures of solid colors, no tessalation/skinning/complex materials etc. and really lowres render target, GI is very low frequency effect.
NVidia’s current VXGI works amazingly well, but it’s really slow in interior scenes for current generation of GPUs (half of the framerate for 8 cones is really bad). Would the screen space method be really that much slower in comparison to VXGI when rendering additional views in a very crude detail and low resolution?
Something i stumbled upong today. Technique has very high quality albeit in presented implementation is very slow (200ms). It works similiar to Geomerics, but it’s fully realtime (do not need to precompute geometry visibility).
Might be interesting to consider it.
),I don’t know if this is any help but I found this 