I would be kinda worried about filtering if applying this to mip generation. I mean, it changes the content significantly.
-
-
-
Trilinear filtering would produce some wrong transitions for sure. I also would be careful whether it makes sense on non-color maps (for normal mapping there are much better approaches, but probably this work could be extended for spectral remapping for glints :) ).
-
But my reference point was game studios using bicubic filter (ok) followed by sharpen (not ok!) because of artists not liking mip generation not preserving look (and “feel”) of original textures.
-
This is clearly better alternative - and spectral remapping is applied only to periodic/residual/non-localized part of signal. Probably can attenuate it if desired :)
-
Yes, bicubic + sharpen is terrible. I think we have to be careful messing with source data too much, though. I can see this sort of technique causing a lot of confusion + frustration on the art side.
-
In our testing Kaiser was better than bicubic plus sharpen.
-
Interesting (non real time) alternative/experiment would be to store set of decomposed Gabor kernels (only decomposition is long and offline process) in a texture and dynamically remap them for continuous zoom-out. Similar to what many glint papers do.
-
(or maybe not so easy/fast - aligning phases is a global optimization step…)
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
Won't you get mush under trilinear interpolation with a frequency "preserving" downsample?
-
Yes, few other folks noticed it (but common game art practice bicubic + sharpen has similar problem). It might be ok but also most console games use trilinear only on part of interpolation transition as optimization (forgot the name, something like 3/8).
-
It seems also like something generally worth exploring for VR where contrast loss due to factors like mipmapping is one of main sources of tunnel vision effect.
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
Outout looks similar to other pattern preserving work. A bit concerned that trilinear interpolation will produce a blurry mess.
-
Generally agree, see my comments I replied to
@JJcoolkl (this approach decomposes signal that is downsampled and remapped, so you can attenuate remapped part if you need to). -
Yeah I saw the rest just after posting
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
-
-
makes sense to use this for the initial downscale from art mastering resolution to in-game max resolution
ধন্যবাদ। আপনার সময়রেখাকে আরো ভালো করে তুলতে টুইটার এটিকে ব্যবহার করবে। পূর্বাবস্থায়পূর্বাবস্থায়
-
-
-
Efficient high quality resampling, even if decimation, is hard
ধন্যবাদ। আপনার সময়রেখাকে আরো ভালো করে তুলতে টুইটার এটিকে ব্যবহার করবে। পূর্বাবস্থায়পূর্বাবস্থায়
-
-
-
This would not work with mip-maps because they would pop and/or blend badly.
-
Yea I was thinking this too. What would the trilinear blend look like as the camera approached a texture with mipmaps computed in this way? It would look odd as it blended between mips, I think. I would have to see it though.
-
The patterns would beat and you would get distracting moire patterns. Also, angled polys would not get texture perspective diminishing, just some kind of repeated mess off into the distance.
-
Yup, angled polys would look odd as mips were blended. I think it's very interesting work, but might not be the best fit for mipmap generation.
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
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।