Thanks, very interesting direction! I'm not entirely sure I understand however how you decide to subdivide - do you only use the final colors of already shaded sparse pixels (and their object IDs)? Or do you scan the G-buffer pixels (e.g. object IDs, or more) inbetween?
-
-
-
In our impl, we shade a finer-level pixel if its already-shaded neighbors have different mtl IDs or the σ² of their (pre-γ) shaded colors exceeds a user-chosen threshold (ref. also footnote pg. 2). Have ideas for better heuristics, but this simple one seems to work well already!
-
This misses details that fall entirely between the coarse level pixels, right?
-
Yes, as long as those details are not accounted for in the heuristic (E.g. mtl ID lets us capture thin geometry; other G-buffer features could be used too (future work)). In-general though, fine features tend to be large-enough, especially as displays become higher resolutions.
-
I wonder if rendertarget compression hardware can be abused to find which 4x4 blocks are smooth and boring and which need subdivision. Probably on consoles?
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
Is there available source code?
-
Implementing DACS merely requires tweaking the shading pass (and removing the checkerboard-reconstruct pass, because it's no longer necessary). Here's some GLSL-like pseudocode for the tweaked shader (plaintext added at original link). Email or DM for clarifications/problems!pic.twitter.com/eW3hFbXRVM
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
Are you aware of this old demoscene production that uses quite a similar approach for raytracing? https://defiance.untergrund.net/vintage/misc/h7/subsample.htm …pic.twitter.com/O8jfB8GnKv
-
Indeed, there are many similar techniques in the ray-tracing literature, which is where we got the idea initially. The subdiv pattern matters a lot—VRay's/our sc 1/⎷2 rot 45° grid captures scene-fn freqs well (while rect subdiv doesn't). We had to omit the comparison for space.
-
If you get to do a presentation at hpg, please add all the juicy details we're missing out on now :)
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
Could this work for Foveated Rendering by using the highest shading rate in the fovea region and fade to the lowest rate in the peripheral vision, instead of rendering 2 or more views per eye with different resolutions (overhead)?
-
Yes, and it works very well! Besides the perf/elegance of simplicity, the shading rate varies smoothly, so you don't have the inter-level discontinuities of existing methods. We removed it from this submission because the reviewers wanted it evaluated more than we had space for.
-
Great, thanks!
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
How does this hold up w.r.t. interpolation artifacts that move location from frame to frame? I don't see that discussed in the paper. If you are starting with a 1/4x1/4 pattern and potentially interpolating 15/16 pixels, I would suspect fair amount of frame-to-frame flicker?
-
Good question. Even only looking at pixel color (UE4 examples) and limited AA/post, the result seems reasonable (see video). Temporal AA should fix lingering issues, as it already does for checkerboard rendering. Earlier revisions discussed all this—but left proof to future work.
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
Read this today, very enjoyable. Good work!
@ian_mallet@cem_yukselধন্যবাদ। আপনার সময়রেখাকে আরো ভালো করে তুলতে টুইটার এটিকে ব্যবহার করবে। পূর্বাবস্থায়পূর্বাবস্থায়
-
-
-
It looks very interesting and certainly sounds promising. Am I reading it correctly that the qualitative analysis was done on videos, but you also did realtime tests? In that case, do you have profiling data from the realtime examples too? (timings, test-hardware, content etc).
ধন্যবাদ। আপনার সময়রেখাকে আরো ভালো করে তুলতে টুইটার এটিকে ব্যবহার করবে। পূর্বাবস্থায়পূর্বাবস্থায়
-
-
-
I met Cem Yuksel briefly at Texas A&M way back in 2010. Nice to see his name come up.
ধন্যবাদ। আপনার সময়রেখাকে আরো ভালো করে তুলতে টুইটার এটিকে ব্যবহার করবে। পূর্বাবস্থায়পূর্বাবস্থায়
-
লোড হতে বেশ কিছুক্ষণ সময় নিচ্ছে।
টুইটার তার ক্ষমতার বাইরে চলে গেছে বা কোনো সাময়িক সমস্যার সম্মুখীন হয়েছে আবার চেষ্টা করুন বা আরও তথ্যের জন্য টুইটারের স্থিতি দেখুন।