Thread: VirtualDub2
View Single Post
Old 22nd October 2018, 03:47   #708  |  Link
WorBry
Registered User
 
Join Date: Jan 2004
Location: Here, there and everywhere
Posts: 1,197
@Shekh,
Sorry, but I'd appreciate further clarification on this:

Quote:
Originally Posted by shekh View Post
Checked what is going on with chroma,
444 -> 422: each pixel is convolved with kernel 0.25, 0.5, 0.25
As I understand this creates some blurring.

422 -> 444: odd pixel is copied as is, even pixel is blended from two neighbor source pixels
This is simple.

Quote:
Originally Posted by shekh View Post
From the screenshot it seems Sony is using bilinear downsampling for 444->422, this blends max 2 neighbor pixels as opposed to summed area.
I changed various interpolation options until I got very similar picture.
For 442->444 it is still bilinear upsampling, no changes.
You said there that VDub2 sub-samples 444 > 422 using this "each pixel convolved with kernel 0.25, 0.5, 0.25" modality, whereas Vegas is using bilinear re-sampling which "blends max 2 neighbor pixels as opposed to summed area".

As it turns out, both Resolve (15.1.2) and Premiere Pro produce exactly the same results as VDub2 when the Checkers-444 (ProRes_4444) clip is 'pass through' exported to v210, giving the same pattern, as per the image in your first post:



Whereas Resolve 15.1 behaves in the same way as Vegas Pro, giving the same pattern as the one in your second post:



Which is also the same as the 'Checkers-422' clip (ProRes_HQ 10bit 422) that was exported from the original image created in Natron. So Natron must use the same sub-sampling modality too:



What's confusing me is that when I use AVISynth+ to do the 444 > 422 conversion with 'bilinear' chroma-resampling and then encode to v210 with VDub2, the pattern of results is exactly the same as when the conversion is done in VDub2 (i.e. importing the original Checkers-444 clip).



Code:
FFVideoSource("{Path}:\checkers-444.mov", fpsnum=25, fpsden=1, threads=1)
Converttoyuv422(interlaced=false, chromaresample="bilinear") # which VDub2 decodes to p210
So I'm left wondering, if your interpretation of the 'Vegas' pattern is 'Bilinear' (by definition) then why is the AVISynth 'Bilinear' re-sampler giving this other pattern that results from "each pixel being convolved with kernel 0.25, 0.5, 0.25 "(summed area) ?

Had a look at the AVISynth 'Resampling' docs to see what I could glean, but I'm none the wiser:

http://avisynth.nl/index.php/Resampl...-_downsampling

Is this other 'summed area' modality a variant of Bilinear or something different altogether ? I've seen mention of a so called 'area averaging' method of sampling (as distinct from 'Bilinear') but only in the context of image scaling algorithms.

https://superuser.com/questions/3757...ose-for-videos

Could you point me to any references that explain, and ideally illustrate this more fully ?

Edit: Here also, but again in relation to resolution scaling:

http://www.enfocus.com/manuals/UserG...ml#c_aa1135518

FFMPEG also:

https://ffmpeg.org/ffmpeg-scaler.htm...Scaler-Options

But nothing I can find relating to 444 > 422 sub-sampling specifically.
__________________
Nostalgia's not what it used to be

Last edited by WorBry; 22nd October 2018 at 04:53.
WorBry is offline   Reply With Quote