How To Downscale 4K 2160p HEVC 10 bit Video To 2K 1080p AVC 10bit

[LouieChuckyMerry]

Happy Friday! and thanks in advance for any help. I'm trying for the first time to downscale a 4K 2060p HEVC 10bit video to 2K 1080p AVC 10bit but am having trouble. The script

# Set DAR in encoder to 1920 : 817. The following line is for automatic signalling
global MeGUI_darx = 1920
global MeGUI_dary = 817
LoadPlugin("PathToLSMASHSource.dll")
LWLibavVideoSource("TestClip[2160p].mkv", format="YUV420P8")
SetFilterMTMode("Default_MT_Mode",2)
SMDegrain(TR=3,ThSAD=500,RefineMotion=True,Plane=0,Chroma=False,n16=True,n16_Out=True)
### Spline36Resize(1920,817).Sharpen(0.2) ###
FastLineDarkenMod4()
F3KDB(Y=100,Cb=100,Cr=100,GrainY=0,GrainC=0)
PreFetch(5)

works as it should, but as soon as I activate the resize line MeGUI crashes. I've spent half a day searching for a solution without any success. I'm thinking it's an issue with the HEVC source but, to be honest, I've no idea .

Here's a short clip[New].

Thanks again for any help .

[davexnet]

Typical delivery format video, with 4:2:0 chroma sampling has to be mod2 sizes, width and height

[Atak_Snajpera]

For better performance I Would move resize line above smdegrain

[LouieChuckyMerry]

Typical delivery format video, with 4:2:0 chroma sampling has to be mod2 sizes, width and height

Thanks for your reply; would you be willing to translate it?

For better performance I Would move resize line above smdegrain

It's my understanding that it's better to denoise before resizing; why would you resize first?

[Atak_Snajpera]

4x Less pixels to denoise. 4x better performance.
Why Would you denoise pixels just to discard them in next step VIA resizing?

[davexnet]

Typical delivery format video, with 4:2:0 chroma sampling has to be mod2 sizes, width and height

Thanks for your reply; would you be willing to translate it?

Translate? It just means width / height must be cleanly divisible by 2, no odd numbers allowed

[Selur]

so, instead of Spline36Resize(1920,817) use either Spline36Resize(1920,818) or Spline36Resize(1920,816)

[LouieChuckyMerry]

4x Less pixels to denoise. 4x better performance.
Why Would you denoise pixels just to discard them in next step VIA resizing?

As explained to me at Doom9 by several people, removing noise first allows for better accuracy on the resize.
_____________

davexnet; Selur: It still crashes with Spline36Resize(1920,818) or Spline36Resize(1920,816).

[davexnet]

4x Less pixels to denoise. 4x better performance.
Why Would you denoise pixels just to discard them in next step VIA resizing?

As explained to me at Doom9 by several people, removing noise first allows for better accuracy on the resize.
_____________

davexnet; Selur: It still crashes with Spline36Resize(1920,818) or Spline36Resize(1920,816).

Does MeGui actually crash with an error code - Or are you getting an Avisynth error message?

[jagabo]

Here's a short clip.

That file is 1920x808, not 4K.

[LouieChuckyMerry]

davexnet: No error code of any kind, MeGUI simply disappears.
_____________

Here's a short clip.

That file is 1920x808, not 4K.

Here's the MediaInfo for the video:

Video 
ID : 1 
Format : HEVC 
Format/Info : High Efficiency Video Coding 
Format profile : Main 10@L5@High 
Codec ID : V_MPEGH/ISO/HEVC 
Duration : 1 h 50 min 
Bit rate : 20.6 Mb/s 
Width : 3 840 pixels 
Height : 1 634 pixels 
Display aspect ratio : 2.35:1 
Frame rate mode : Constant 
Frame rate : 23.976 FPS 
Color space : YUV 
Chroma subsampling : 4:2:0 
Bit depth : 10 bits 
Bits/(Pixel*Frame) : 0.137 
Stream size : 15.9 GiB (85%) 
Default : Yes 
Forced : No 
Color range : Limited 
Color primaries : BT.709 
Transfer characteristics : BT.709 
Matrix coefficients : BT.709

[Alwyn]

@LouieChuckMerry, the video on your "Here's a short clip" link in your post #1 is as Jagabo says:

General
Unique ID                                : 298982791378727529867026321803490179530 (0xE0EDFDB73D8DEB7D37532C65EA2D39CA)
Complete name                            : C:\Users\bb19\Downloads\TestClip[2160p].mkv
Format                                   : Matroska
Format version                           : Version 4
File size                                : 14.4 MiB
Duration                                 : 5 s 89 ms
Overall bit rate                         : 23.8 Mb/s
Frame rate                               : 23.976 FPS
Encoded date                             : 2023-05-05 19:28:48 UTC
Writing application                      : mkvmerge v68.0.0 ('The Curtain') 64-bit
Writing library                          : libebml v1.4.2 + libmatroska v1.6.4

Video
ID                                       : 1
Format                                   : AVC
Format/Info                              : Advanced Video Codec
Format profile                           : High@L4.1
Format settings                          : CABAC / 5 Ref Frames
Format settings, CABAC                   : Yes
Format settings, Reference frames        : 5 frames
Codec ID                                 : V_MPEG4/ISO/AVC
Duration                                 : 5 s 89 ms
Bit rate                                 : 23.7 Mb/s
Width                                    : 1 920 pixels
Height                                   : 808 pixels
Display aspect ratio                     : 2.40:1
Frame rate mode                          : Constant
Frame rate                               : 23.976 (23976/1000) FPS
Color space                              : YUV
Chroma subsampling                       : 4:2:0
Bit depth                                : 8 bits
Scan type                                : Progressive

Is it not the HEVC file you reference in the post above.

[davexnet]

I would create a manual script in Notepad, and see if it opens in VIrtualdub2 or AVSPmod
See if it performs as expected.
As was mentioned previously, the clip posted in the first post above, is not what you're describing.
It's neither HEVC, 10-bit nor 4K

[LouieChuckyMerry]

ALL: I'm really sorry and not sure why my uploaded clip is not as the source (I did wonder why a five second section cut from a source using MKVToolNix registered as a five-plus minutes file in MediaInfo). I'll make a more accurate clip as soon as I can, thanks for your patience.

[Atak_Snajpera]

4x Less pixels to denoise. 4x better performance.
Why Would you denoise pixels just to discard them in next step VIA resizing?

As explained to me at Doom9 by several people, removing noise first allows for better accuracy on the resize.
_____________

davexnet; Selur: It still crashes with Spline36Resize(1920,818) or Spline36Resize(1920,816).

That's total BS. Simple Spline36Resize filter does not need denoised pixels to be accurate. Besides when you downscale you already perform denoising because 4 pixels will be averaged to just 1 pixel. What you do is extremely inefficient. Your time , your electricity bill and your placebo effect...

[Selur]

That's total BS.

It's not.
It's easier/more accurate to properly clean (denoise, deblock, derainbow,..) in the native resolution, especially if you upscale afterward.
If you !downscale!, depending on the noise and artifacts in the source, it can be fine to resize first, and it will be beneficial to the processing speed.

[Atak_Snajpera]

But we are NOT upscaling! We are downscaling. No need to waste time for pixels which will be discarded anyway and then butchered by H.264/AVC compression algorithms.

[Selur]

True, but the general advice isn't BS. + depending on the nature of the source, downscaling can unnecessarily change the nature of artifacts&co and thus hinder proper cleaning.
So your advice isn't really that good of a general advice either.
=> it always depends

[Atak_Snajpera]

it is obvious what I meant if you read his post

As explained to me at Doom9 by several people, removing noise first allows for better accuracy on the resize.

[Selur]

Not, to me. To me, that seems more like a not so well phrased statement. And it's more about the quality of the resize than the accuracy.
So I agree, to disagree and think your blatant

That's total BS.

still is unwarranted.

[lollo]

I agree with Selur. Saying "That's total BS." is wrong.

If you denoise before upscale is always better. If you denoise before downscale it can produce better results or it can be transparent with respect to a denoise after downscale.

Then, if you denoise before you're never wrong, and it is always the best blind choice. Eventually, you may loose resources efficiency and electricity bill in downscaling under certain conditions, but is marginal and neglettable.

[Atak_Snajpera]

When you downscale from 4K to FHD you are averaging 4 pixels to 1 pixel.
Example



This means that you do not have to use aggressive denoising settings later in the script. More aggressive settings in SMdegrain may introduce ghosting artefacts. Simple!

[lollo]

If one if the four gray pixel is noise, it will be included in the average pixel while it should be not, and it will be not possible to remove it later after downscale.
There are other cases where this do not apply.
Simple!

P.S. SMdegrain perform its best feature as a temporal denoiser, meaning that it looks to denoise between frames and not inside same frame, like TemporalDegrain2. Choose a pure spatial denoiser for your example

[Atak_Snajpera]

You live in perfect world where SMDegrain can ALWAYS detect noise perfectly. Averaged pixels will be less visible to our eyes so what is your problem? Besides lossy codecs like x264/x265 will destroy your fine pixels anyways by further "averaging" those pixels.

Stop living in placebo bubble and do not waste energy.

[lollo]

so what is your problem?

My problem is that you are technically wrong and instead replying with facts bring into the discussion "points of view"

Besides x264/x265 will destroy your fine pixels anyways by further "averaging" those pixels.

I agree. But we are talking about science before everything else. Then, given the further compression, we may consider that any video improvement is a waste of time (but at low CRF even h264 is not really different from original at normal view, so it is worth, instead)

Stop living in placebo world and do not waste energy.

There is no waste of enery trying to obtain the best possible result. Sometimes the chosen road can be not the shortest, but what is important is the final results. Reread my post #21 and you'll understand better.

[Atak_Snajpera]

Have you ever heard about term "Diminishing returns"
https://en.wikipedia.org/wiki/Diminishing_returns

I'm talking about common sense. You just like placebo effect.

[lollo]

Have you ever heard about term "Diminishing returns"

I am familiar with any concept you can bring into our discussion, yes. Which does not change a comma in what said.

I'm talking about common sense. You just like placebo effect.

It's just your conclusion. (You didn't re-read post number 21, did you?)

[LouieChuckyMerry]

ALL: I used Avidemux to cut the clip and it seems OK; the link in my original post has been edited.
_____________

I would create a manual script in Notepad, and see if it opens in VIrtualdub2 or AVSPmod
See if it performs as expected.
As was mentioned previously, the clip posted in the first post above, is not what you're describing.
It's neither HEVC, 10-bit nor 4K

This script:

# Set DAR in encoder to 1920 : 816. The following line is for automatic signalling
global MeGUI_darx = 1920
global MeGUI_dary = 816
LoadPlugin("PathToLSMASHSource.dll")
LWLibavVideoSource("TestClip[2160p].mkv", format="YUV420P8")
SetFilterMTMode("Default_MT_Mode",2)
SMDegrain(TR=3,ThSAD=500,RefineMotion=True,Plane=0,Chroma=False,n16=True,n16_Out=True)
Spline36Resize(1920,816).Sharpen(0.2) ###
FastLineDarkenMod4()
F3KDB(Y=100,Cb=100,Cr=100,GrainY=0,GrainC=0)
PreFetch(5)

Opens in both VirtualDub2 and AvsPmod. I also noticed that when adding the source to MeGUI for indexing, the option to resize is grayed out. Is it possibly an MeGUI issue? Maybe I'll resize it in VirtualDub2 using a lossless codec then clean it up and shrivel it with MeGUI.
_____________

Atak_Snajpera: For what it's worth, when I began trying to learn AviSynth I wanted to know the "best" way to order things and wound up at Doom9. After much helpful input I settled on:

# Deinterlace
# Color Conversion
# Color Adjustment
# Crop
# Denoise
# Resize
# Sharpen
# Deband
# Dither

Of course, that was going on ten years ago so maybe things have changed since then; I'm far far far from expert so I've just stuck with the above because it seems reasonable to me .

[jagabo]

Maybe that "global MeGUI_dary = 817" is a problem in MeGUI?

[Selur]

Try mClean instead of SMDegrain.

Cu Selur

Ps.: SCUNet, isn't bad to remove such noise: https://imgsli.com/MTc2NTM4 (but it's probably too slow and would require Vapoursynth)