Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Capturing and Editing Video > Avisynth Development

Reply
 
Thread Tools Search this Thread Display Modes
Old 11th January 2022, 12:08   #1761  |  Link
DTL
Registered User
 
Join Date: Jul 2018
Posts: 342
It is good to add to wiki about RGB and YUV formats (http://avisynth.nl/index.php/Color_spaces ): the digital YUV colorspace is more limited in precision in compare with digital RGB colorspace. So for comparable quality it is about RGB 8bit and YUV 10bit. One consequence mean using YUV8 (YUV 4xx 8bit, typical storage) stage for converting to RGB8 (RGB24, typical display) cause more digital distortions (banding).

YUV cover 'wider' colour gamut in compare with RGB but this cause precision lost with 'natural' colour gamut with same storage integer bits per value. So the conversions RGB<->YUV with same bitdepth are not lossless even with YUV444. The error is about 1 LSB but it visible on smooth colour gradients with 8bit. At first I think it is an issue in Convert() functions.
May be it is already somewhere covered in AVS wiki ?

Last edited by DTL; 11th January 2022 at 12:20.
DTL is offline   Reply With Quote
Old 11th January 2022, 22:41   #1762  |  Link
LigH
German doom9/Gleitz SuMo
 
LigH's Avatar
 
Join Date: Oct 2001
Location: Germany, rural Altmark
Posts: 6,501
Your conclusion is not completely correct. The precision reduction has two possible reasons: "TV range" (but full range YUV exists too) and float arithmetics in the YCbCr conversion formulas (avoidable in the YCgCo model). If you don't need to convert between RGB and YUV, then full range YUV has the same precision of 8 bits per component as RGB. Thus, YUV is not generally worse; it is only worse for RGB based devices.
__________________

New German Gleitz board
MediaFire: x264 | x265 | VPx | AOM | Xvid
LigH is offline   Reply With Quote
Old 12th January 2022, 01:17   #1763  |  Link
DTL
Registered User
 
Join Date: Jul 2018
Posts: 342
""TV range" (but full range YUV exists too)"

Here is the test: taking RGB $808080 something around mid-grey colour (no-colour grey) and slowly increase saturation of green
Code:
BlankClip(color=$FF808080)

ConvertToYUV444(matrix="PC.601")
ConvertToRGB32(matrix="PC.601")

ScriptClip ("""Subtitle (String (AverageR()) + " " + String (AverageG()) +" "+ String (AverageB()))""")
Input $FF808080 - out 128 128 128 = OK
Input $FF808180 - out 129 129 129 < ??? - luma raised, no green colour = error in colour tone and colour saturatrion
Input $FF808280 - out 128 130 127 < ??? - green colour + some minus-blue colour = error in colour tone
Input $FF808380 - out 129 131 128 < ??? - green colour + some plus-red colour = error in colour tone

The only known way to fix - use >8 bit processing of RGB->YUV->RGB:
Code:
BlankClip(color=$FF808080)

ConvertBits(16)

ConvertToYUV444(matrix="PC.601")
ConvertToRGB32(matrix="PC.601")

ConvertBits(8)

ScriptClip ("""Subtitle (String (AverageR()) + " " + String (AverageG()) +" "+ String (AverageB()))""")
Input $FF808080 - out 128 128 128 = OK
Input $FF808180 - out 128 129 128 = OK
Input $FF808280 - out 128 130 128 = OK
Input $FF808380 - out 128 131 128 = OK

Last edited by DTL; 12th January 2022 at 01:25.
DTL is offline   Reply With Quote
Old 12th January 2022, 02:54   #1764  |  Link
poisondeathray
Registered User
 
Join Date: Sep 2007
Posts: 4,802
Quote:
Originally Posted by DTL View Post
It is good to add to wiki about RGB and YUV formats (http://avisynth.nl/index.php/Color_spaces ): the digital YUV colorspace is more limited in precision in compare with digital RGB colorspace. So for comparable quality it is about RGB 8bit and YUV 10bit. One consequence mean using YUV8 (YUV 4xx 8bit, typical storage) stage for converting to RGB8 (RGB24, typical display) cause more digital distortions (banding).

YUV cover 'wider' colour gamut in compare with RGB but this cause precision lost with 'natural' colour gamut with same storage integer bits per value. So the conversions RGB<->YUV with same bitdepth are not lossless even with YUV444. The error is about 1 LSB but it visible on smooth colour gradients with 8bit. At first I think it is an issue in Convert() functions.
May be it is already somewhere covered in AVS wiki ?



That is the expected result for 8bit RGB=>YUV=>RGB round trip. +/- 3 value errors

Not just precision and 8bit rounding errors, but multiple 8bit YUV values can map to the same RGB value, which produces more "banding" instead of smooth gradient

In general, 10bit YUV or greater is required for lossless round trip from 8bit RGB => YUV => 8bit RGB, if implemented properly.


But you can demonstrate cases where 10bit is not sufficient with internal Convert(), yet 10bit is sufficient with AvsResize/zimg. ( 10bit also sufficient with other programs like NLE's) .

In this example, 12bit with internal convert is required , but 10bit with AvsResize/zimg or other programs works ok
https://forum.doom9.org/showthread.p...86#post1897686

This is pinterf's explanation for the difference using internal convert
https://forum.doom9.org/showthread.p...83#post1902783
poisondeathray is offline   Reply With Quote
Old 12th January 2022, 08:41   #1765  |  Link
LigH
German doom9/Gleitz SuMo
 
LigH's Avatar
 
Join Date: Oct 2001
Location: Germany, rural Altmark
Posts: 6,501
(matrix="PC.601") ... please read: Wikipedia: YUV, specifically chapters 2.1: SDTV with BT.470 (oh, look, fractional matrix coefficients with 3 decimal places) and 3: Numerical approximations (okay, small integers here, but with rounding errors, and they cause most of the banding, especially because they need to be applied forth and back).

A YCoCg matrix has much simpler coefficients, halves and quarters, nice to handle in binary form.
__________________

New German Gleitz board
MediaFire: x264 | x265 | VPx | AOM | Xvid
LigH is offline   Reply With Quote
Old 12th January 2022, 14:51   #1766  |  Link
DTL
Registered User
 
Join Date: Jul 2018
Posts: 342
"3: Numerical approximations (okay, small integers here, but with rounding errors, and they cause most of the banding, especially because they need to be applied forth and back)."

May be good to copy to Avisynth wiki (and may be Convert() functions note). May be possible to develop some 'pre-distortions' for 8bit format conversions to minimize colour tone shift (that is more visible ?) ? Currently the distribution of rounding errors is equal for colour tone, luma and saturation errors ? I see typical solutions is add dithering. Unfortunately 8bit still widely used and in end-users displays it is the only supported with h.264 encoding. So using of h.265 with 10bit decreases number of possible users of encoded content.
DTL is offline   Reply With Quote
Old 14th January 2022, 08:39   #1767  |  Link
pinterf
Registered User
 
Join Date: Jan 2014
Posts: 2,113
Avisynth 3.7.2 test 1
Avisynth+ 3.7.2 test 1 (20220113)

Code:
20220113 3.7.2-WIP
------------------
- Fix: Attempt to resolve deadlock when an Eval'd (Prefetch inside) Clip result is 
  used in Invoke which calls a filter with GetFrame in its constructor.
  (AvsPMod use case which Invokes frame prop read / ConvertToRGB32 after having the AVS script evaluated)
  Remark: problem emerged in 3.7.1test22 which is trying to read frame properties of the 0th frame in its constructor.
  A similar deadlock situation was already fixed earlier in Neo branch and had been backported but it did not cover this use case.
- Fix: Histogram AudioLevels half character upshift (regression since v3.6)
- Bump Copyright year to 2022
pinterf is offline   Reply With Quote
Old 14th January 2022, 11:06   #1768  |  Link
FranceBB
Broadcast Encoder
 
FranceBB's Avatar
 
Join Date: Nov 2013
Location: Metropolitan City of Milan, Italy
Posts: 1,933
Nice one, Ferenc!
But... it doesn't work.

Quote:
Originally Posted by pinterf View Post
Remark: problem emerged in 3.7.1test22 which is trying to read frame properties of the 0th frame in its constructor.
the problem is still there I'm afraid...


With 3.7.1 Stable:



And I had to get rid of frame properties to make it work:




With 3.7.2 Test 1 it's the same thing:



and once I get rid of frame properties...



Source:

Code:
General 
Complete name : \\mibctvan000.avid.mi.bc.sky.it\Ingest\MEDIA\temp\UCN12982_GHOST_FRAME.mxf 
Format : MXF 
Commercial name : XDCAM HD422 
Format version : 1.3 
Format profile : OP-1a 
Format settings : Closed / Complete 
File size : 5.50 GiB 
Duration : 14 min 58 s 
Overall bit rate : 52.6 Mb/s 
Package name : Source Package 
Encoded date : 2022-01-12 10:16:12.076 
Writing application : Avid Technology, Inc. Avid MediaProcessor Plug-In 1.0.54.10052.1 
Writing library : MXF::SDK (4.7.8) on Win64 4.7.8.10137.1 

Video 
ID : 512 
Format : MPEG Video 
Commercial name : XDCAM HD422 
Format version : Version 2 
Format profile : 4:2:2@High 
Format settings : CustomMatrix / BVOP 
Format settings, BVOP : Yes 
Format settings, Matrix : Custom 
Format settings, GOP : Variable 
Format settings, picture structure : Frame 
Format settings, wrapping mode : Frame 
Codec ID : 0D01030102046001-0401020201040300 
Duration : 14 min 58 s 
Bit rate mode : Constant 
Bit rate : 50.0 Mb/s 
Width : 1 920 pixels 
Height : 1 080 pixels 
Display aspect ratio : 16:9 
Frame rate : 25.000 FPS 
Standard : PAL 
Color space : YUV 
Chroma subsampling : 4:2:2 
Bit depth : 8 bits 
Scan type : Interlaced 
Scan order : Top Field First 
Compression mode : Lossy 
Bits/(Pixel*Frame) : 0.965 
Time code of first frame : 00:00:00:00 
Time code source : Group of pictures header 
Stream size : 5.23 GiB (95%) 
Color range : Limited 
Color primaries : BT.709 
Transfer characteristics : BT.709 
Matrix coefficients : BT.709 

Audio #1 
ID : 768 
Format : PCM 
Format settings : Little 
Format settings, wrapping mode : Frame (AES) 
Codec ID : 0D01030102060300-0402020101000000 
Duration : 14 min 58 s 
Bit rate mode : Constant 
Bit rate : 1 152 kb/s 
Channel(s) : 1 channel 
Sampling rate : 48.0 kHz 
Frame rate : 25.000 FPS (1920 SPF) 
Bit depth : 24 bits 
Stream size : 123 MiB (2%) 
Locked : Yes 

Audio #2 
ID : 1024 
Format : PCM 
Format settings : Little 
Format settings, wrapping mode : Frame (AES) 
Codec ID : 0D01030102060300-0402020101000000 
Duration : 14 min 58 s 
Bit rate mode : Constant 
Bit rate : 1 152 kb/s 
Channel(s) : 1 channel 
Sampling rate : 48.0 kHz 
Frame rate : 25.000 FPS (1920 SPF) 
Bit depth : 24 bits 
Stream size : 123 MiB (2%) 
Locked : Yes 

Other #1 
ID : 1-Material 
Type : Time code 
Format : MXF TC 
Frame rate : 25.000 FPS 
Time code of first frame : 10:00:00:00 
Time code settings : Material Package 
Time code, striped : Yes 
Title : Timecode 

Other #2 
ID : 0-Source 
Type : Time code 
Format : MXF TC 
Frame rate : 25.000 FPS 
Time code of first frame : 10:00:00:00 
Time code settings : Source Package 
Time code, striped : Yes 

Other #3 
Type : Time code 
Format : SMPTE TC 
Muxing mode : SDTI 
Frame rate : 25.000 FPS 
Time code of first frame : 10:00:00:00
Disabling "Read matrix from source or script" in AVSPmod fixes it, but still...

Using VirtualDub, it works, regardless of frame properties:

FranceBB is online now   Reply With Quote
Old 14th January 2022, 11:29   #1769  |  Link
pinterf
Registered User
 
Join Date: Jan 2014
Posts: 2,113
This issue must be different, there is no Prefetch here.
EDIT:
Can you put a ConvertToRGB32() at end of the script?
(to check if it fails; AvsPMod (as far as I saw in the source) does two additional things after evaluating our original AVS script: reads frame properties and converts the clip to rgb32 in order to display it.


My fix was intended to heal only the "hang" issue.

Last edited by pinterf; 14th January 2022 at 14:33.
pinterf is offline   Reply With Quote
Old 14th January 2022, 17:41   #1770  |  Link
gispos
Registered User
 
Join Date: Oct 2018
Location: Germany
Posts: 635
As soon as FranceBB has it in his hands, he breaks it.

Try this version, no problems for me.

32bit:
https://www.mediafire.com/file/ar8qp...6-32).zip/file
64bit:
https://www.mediafire.com/file/uvtp8...6-64).zip/file
__________________
Live and let live
gispos is offline   Reply With Quote
Old 14th January 2022, 18:25   #1771  |  Link
FranceBB
Broadcast Encoder
 
FranceBB's Avatar
 
Join Date: Nov 2013
Location: Metropolitan City of Milan, Italy
Posts: 1,933
Quote:
Originally Posted by gispos View Post
Nope.
Same error as before.
Please note that this happens only with XDCAM files muxed in .mxf
I can share a sample if you want, 'cause I've just tried with a ProRes file muxed in .mov and it works just fine.
Same goes for an H.264 file in .mp4.

Quote:
Originally Posted by gispos View Post
As soon as FranceBB has it in his hands, he breaks it.
eheheheh nah, it's more like: "literally no one except FranceBB uses weird FULL HD yv16 50 Mbit/s 25i TFF MPEG-2 files muxed in mxf anymore, so no one ever checked" xD

Quote:
Originally Posted by pinterf View Post
Can you put a ConvertToRGB32() at end of the script?

Interesting error, it fails with this:



however this is a bit weird given that it's a simple yv16 with the standard MPEG-2 chroma placement (after all it IS an MPEG-2 eheheheh).

Adding propclearall() fixes it and allows me to use ConverttoRGB32():

Code:
video=LWLibavVideoSource("\\mibctvan000.avid.mi.bc.sky.it\Ingest\MEDIA\temp\UCN12982_GHOST_FRAME.mxf")
ch1=LWLibavAudioSource("\\mibctvan000.avid.mi.bc.sky.it\Ingest\MEDIA\temp\UCN12982_GHOST_FRAME.mxf", stream_index=1)
ch2=LWLibavAudioSource("\\mibctvan000.avid.mi.bc.sky.it\Ingest\MEDIA\temp\UCN12982_GHOST_FRAME.mxf", stream_index=2)
audio=MergeChannels(ch1, ch2, ch1, ch2, ch1, ch2, ch1, ch2)
AudioDub(video, audio)

propClearAll()

ConverttoRGB32()
speaking of which, let me see if I can just get rid of the chroma location property and try with this:

Code:
video=LWLibavVideoSource("\\mibctvan000.avid.mi.bc.sky.it\Ingest\MEDIA\temp\UCN12982_GHOST_FRAME.mxf")
ch1=LWLibavAudioSource("\\mibctvan000.avid.mi.bc.sky.it\Ingest\MEDIA\temp\UCN12982_GHOST_FRAME.mxf", stream_index=1)
ch2=LWLibavAudioSource("\\mibctvan000.avid.mi.bc.sky.it\Ingest\MEDIA\temp\UCN12982_GHOST_FRAME.mxf", stream_index=2)
audio=MergeChannels(ch1, ch2, ch1, ch2, ch1, ch2, ch1, ch2)
AudioDub(video, audio)

propDelete("_ChromaLocation")

Success!
Getting rid of the Chroma Location property makes it work!




So now we know what happened!
Basically ConverttoRGB32() fails due to the Chroma Location frame property, therefore when I try to use AVSPmod, it tries to convert to RGB32 invoking such a function, but since such a function fails, it won't display anything and fail instead!!
FranceBB is online now   Reply With Quote
Old 14th January 2022, 18:38   #1772  |  Link
pinterf
Registered User
 
Join Date: Jan 2014
Posts: 2,113
Thanks! Can you put a propShow() before the property delete, to see what the original _ChromaLocation was set to?
pinterf is offline   Reply With Quote
Old 14th January 2022, 18:50   #1773  |  Link
gispos
Registered User
 
Join Date: Oct 2018
Location: Germany
Posts: 635
Quote:
Originally Posted by pinterf View Post
(to check if it fails; AvsPMod (as far as I saw in the source) does two additional things after evaluating our original AVS script: reads frame properties and converts the clip to rgb32 in order to display it.
The clip itself is not converted to RGB32. A DisplayClip is created which is converted to RGB32.
But the frame properties (matrix) are read from the original source clip.
__________________
Live and let live
gispos is offline   Reply With Quote
Old 14th January 2022, 18:55   #1774  |  Link
FranceBB
Broadcast Encoder
 
FranceBB's Avatar
 
Join Date: Nov 2013
Location: Metropolitan City of Milan, Italy
Posts: 1,933
Quote:
Originally Posted by pinterf View Post
Thanks! Can you put a propShow() before the property delete, to see what the original _ChromaLocation was set to?
Sure thing, there you go:



Looks like the indexer is reporting another kind of Chroma Location, 'cause it should be _ChromaLocation = 0 left (mpeg2) and not _ChromaLocation = 2 top_left... interesting...

This is another XDCAM file and it has the very same problem:



and ffprobe seems to agree with the indexers:

[STREAM]
index=0
codec_name=mpeg2video
codec_long_name=MPEG-2 video
profile=4:2:2
codec_type=video
codec_tag_string=[0][0][0][0]
codec_tag=0x0000
width=1920
height=1080
coded_width=0
coded_height=0
closed_captions=0
film_grain=0
has_b_frames=1
sample_aspect_ratio=1:1
display_aspect_ratio=16:9
pix_fmt=yuv422p
level=2
color_range=tv
color_space=bt709
color_transfer=bt709
color_primaries=bt709
chroma_location=topleft

This is yet a different XDCAM file I received from A&E:

index=0
codec_name=mpeg2video
codec_long_name=MPEG-2 video
profile=4:2:2
codec_type=video
codec_tag_string=[0][0][0][0]
codec_tag=0x0000
width=1920
height=1080
coded_width=0
coded_height=0
closed_captions=0
film_grain=0
has_b_frames=1
sample_aspect_ratio=1:1
display_aspect_ratio=16:9
pix_fmt=yuv422p
level=2
color_range=tv
color_space=unknown
color_transfer=bt709
color_primaries=unknown
chroma_location=topleft

and this is a movie I've got from Notorious Pictures and it says top left too:

[STREAM]
index=0
codec_name=mpeg2video
codec_long_name=MPEG-2 video
profile=4:2:2
codec_type=video
codec_tag_string=[0][0][0][0]
codec_tag=0x0000
width=1920
height=1080
coded_width=0
coded_height=0
closed_captions=0
film_grain=0
has_b_frames=1
sample_aspect_ratio=1:1
display_aspect_ratio=16:9
pix_fmt=yuv422p
level=2
color_range=tv
color_space=unknown
color_transfer=bt709
color_primaries=unknown
chroma_location=topleft


Ok, either there's something fishy and totally broken here, or I've lost my certainties...
FranceBB is online now   Reply With Quote
Old 14th January 2022, 19:03   #1775  |  Link
FranceBB
Broadcast Encoder
 
FranceBB's Avatar
 
Join Date: Nov 2013
Location: Metropolitan City of Milan, Italy
Posts: 1,933
Ok, guys, for the glory and my mental health, what's the REAL chroma location of this file: https://we.tl/t-MhjJWTj0lQ

and is the metadata just wrong or is it FFMpeg and the indexers that are not getting it right?

(link available for 7 days)
FranceBB is online now   Reply With Quote
Old 14th January 2022, 19:06   #1776  |  Link
pinterf
Registered User
 
Join Date: Jan 2014
Posts: 2,113
Topleft is treated as invalid for 422, it's accepted only for 420
pinterf is offline   Reply With Quote
Old 14th January 2022, 19:16   #1777  |  Link
FranceBB
Broadcast Encoder
 
FranceBB's Avatar
 
Join Date: Nov 2013
Location: Metropolitan City of Milan, Italy
Posts: 1,933
Quote:
Originally Posted by pinterf View Post
Topleft is treated as invalid for 422, it's accepted only for 420
and it makes sense, it shouldn't exist for 4:2:2, like, I've never ever seen it.
I mean, let's think this through logically.

Top Left is Type 2, so the one generally found in H.265 UHD HDR PQ BD.

Now, those files are MPEG-2 50 Mbit/s BT709 SDR in FULL HD and 4:2:2 8bit planar.
Sony made the XDCAM standard in 2003 and the XDCAM-50 is from 2006 or something like that if I remember correctly.
Back then, Type 2 top left didn't even exist, did it?
Besides, I would find hard to believe that such a standard uses top left instead of left, given that it's an MPEG-2, right?

So we can assume that indexers and ffprobe are all wrong and that it's actually Type 0, left, MPEG-2, right?

If someone is willing to download the masterfile and check I would really appreciate it.
Also 'cause if it's actually Type 0 and indexers + FFMpeg say "Type 2", this is a BIG BIG PROBLEM and we need a fix soon.
FranceBB is online now   Reply With Quote
Old 14th January 2022, 22:18   #1778  |  Link
Dogway
Registered User
 
Join Date: Nov 2009
Posts: 1,637
Same error trying to display the clip, I'm stable 3.7.1 and AvspMod 2.6.7.9 with "read from source script" disabled. propclearall() fixes it. I will try to check chroma placement, haven't seen an empirical method to find out so.

EDIT: Yes, looks MPEG2 to me
__________________
[i7-4790K@Stock::GTX 1070] AviSynth+ filters and mods on GitHub

Last edited by Dogway; 15th January 2022 at 01:22.
Dogway is offline   Reply With Quote
Old 15th January 2022, 05:17   #1779  |  Link
takla
Registered User
 
Join Date: May 2018
Posts: 58
Quote:
Originally Posted by FranceBB View Post
and it makes sense, it shouldn't exist for 4:2:2, like, I've never ever seen it.
I mean, let's think this through logically.

Top Left is Type 2, so the one generally found in H.265 UHD HDR PQ BD.

Now, those files are MPEG-2 50 Mbit/s BT709 SDR in FULL HD and 4:2:2 8bit planar.
Sony made the XDCAM standard in 2003 and the XDCAM-50 is from 2006 or something like that if I remember correctly.
Back then, Type 2 top left didn't even exist, did it?
Besides, I would find hard to believe that such a standard uses top left instead of left, given that it's an MPEG-2, right?

So we can assume that indexers and ffprobe are all wrong and that it's actually Type 0, left, MPEG-2, right?

If someone is willing to download the masterfile and check I would really appreciate it.
Also 'cause if it's actually Type 0 and indexers + FFMpeg say "Type 2", this is a BIG BIG PROBLEM and we need a fix soon.
Maybe get in touch with ffmpeg devs?
https://ffmpeg.org/contact.html#IRCChannels
https://ffmpeg.org/contact.html#MailingLists
takla is offline   Reply With Quote
Old 15th January 2022, 11:42   #1780  |  Link
FranceBB
Broadcast Encoder
 
FranceBB's Avatar
 
Join Date: Nov 2013
Location: Metropolitan City of Milan, Italy
Posts: 1,933
Quote:
Originally Posted by takla View Post
I opened a ticket: https://trac.ffmpeg.org/ticket/9598#ticket
FranceBB is online now   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 21:02.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, vBulletin Solutions Inc.