Weird combing in progressive source

[embis2003]

I have a H264 encode of a music video from VEVO. This encode is sourced from the Universal Music Group master file. Apparently, someone at UMG or VEVO didn't deinterlace BEFORE resizing (my bets are on UMG because THIS master had the same problem) which leaves these horrible teeth or combing artifacts all over the video, but they are especially visible during fades, transitions, or scene changes. Perhaps this is a different phenomena then the post referenced above entirely and if so feel free to enlighten me, but either way, these artifacts are a pain and its the only thing making this video stand out from my collection. I was told before to use svpflow to interpolate frames. I'll take blending over combing anyday. It worked in some cases, however, often the artifacts would appear in sequences of 3 or more frames or the frame would be right before a scene change so it would it simply duplicate the frame after it, which ended up looking more noticeable then the combing. Is there anything, and I mean anything at all, that I can do remove or if not reduce the appearance of the combing? At this point, I am willing to pay to have the artifacts removed.

Examples:


[Attachment 52957 - Click to enlarge]


[Attachment 52954 - Click to enlarge]


[Attachment 52955 - Click to enlarge]

Video:

https://drive.google.com/open?id=1ASYICq9-e-mCx1sWrtIKnnrTiZ3nrCI0

[ProWo]

These are intentional video fades for me.

[embis2003]

These are intentional video fades for me.

Perhaps you don't see the combing. I know that the fades are intentional... I changed the first picture to a better example. The second one its on the girl on the swing.

[johnmeyer]

I see no issues with the video you posted.

[embis2003]

I see no issues with the video you posted.

Wh... What? How is that even possible? Did you see the artifacts in the screenshots above? Are those not present when you play the video back? If not, then what in the world are using to play it back? I'm very confused lol

[DB83]

Oh goody !!! Here we go again. LOL !!!!!!!!


Any chance of pointing to the original VEVO/yt/any other source ?


Just curious how a US Record Label ends up with a 25fps 'Master'


There are artifacts. Maybe johnmeyer was viewing the GoogleDrive playback rather than a downloaded version.

[jagabo]

The cure is probably going to be worse than the problem. To get rid of those comb artifacts will require downsizing to a height that blurs them away (about 160 lines) then upscale back to full height. That will leave you with a blurry picture and blend deinterlacing artifacts. Since it's only a few frames here and there you can try using a smaller height (a size that results in near normal comb artifacts) as a test clip with the runtime filter IsCombedIVTC:

LSmashVideoSource("1.mp4") 
patch = Spline36Resize(width, 160).Spline36Resize(width, height).Subtitle("patch") # subtitle to make the patched frames stand out.
testclip = Spline36Resize(1920,328).Sharpen(0.0, 1.0)
ConditionalFilter(testclip, patch, last, "IsCombedTIVTC")
#Overlay(last, testclip, y=32) # for debugging, so you can see the testclip

You can try using different heights and sharpening for testclip. Enable the Overlay() line to show the testclip over the resulting video. That caught most of the frames with residual interlacing. It also screwed up a few frames without interlacing because something in the picture fell in the right frequency range. You might also try using a mask of where the comb artifacts are to limit the patching to just those areas.


[Attachment 52975 - Click to enlarge]

The video probably also needs decimation down to 24p.

[johnmeyer]

There are artifacts. Maybe johnmeyer was viewing the GoogleDrive playback rather than a downloaded version.

Yes, I was.

[embis2003]

Oh goody !!! Here we go again. LOL !!!!!!!!


Any chance of pointing to the original VEVO/yt/any other source ?


Just curious how a US Record Label ends up with a 25fps 'Master'


There are artifacts. Maybe johnmeyer was viewing the GoogleDrive playback rather than a downloaded version.

We meet again.

Any chance of pointing to the original VEVO/yt/any other source ?

Here is the VEVO file: http://embed.vevo.com/?video=USUV70502343

Just curious how a US Record Label ends up with a 25fps 'Master'

That's a great question... but its the reality if you rip it from the link above.

There are artifacts. Maybe johnmeyer was viewing the GoogleDrive playback rather than a downloaded version.

I can still see them very clearly in the preview version too.

FYI, I'm not saying THIS file is a master, its far from it. But the file that it was converted from IS. Because it is the same exact file as https://www.youtube.com/watch?v=1wYNFfgrXTI, which is evident by combing in the same places, however, strangely less visible. The only difference between this and the YouTube version is it is upscaled and is of a higher bitrate.

Also, given that it seems nearly evey master file UMG has contains some sort of combing, its more proof that our old thread was indeed a master. And that it wasn't whoever encoded that file who screwed it up.

[embis2003]

The cure is probably going to be worse than the problem. To get rid of those comb artifacts will require downsizing to a height that blurs them away (about 160 lines) then upscale back to full height. That will leave you with a blurry picture and blend deinterlacing artifacts. Since it's only a few frames here and there you can try using a smaller height (a size that results in near normal comb artifacts) as a test clip with the runtime filter IsCombedIVTC:

LSmashVideoSource("1.mp4") 
patch = Spline36Resize(width, 160).Spline36Resize(width, height).Subtitle("patch") # subtitle to make the patched frames stand out.
testclip = Spline36Resize(1920,328).Sharpen(0.0, 1.0)
ConditionalFilter(testclip, patch, last, "IsCombedTIVTC")
#Overlay(last, testclip, y=32) # for debugging, so you can see the testclip

You can try using different heights and sharpening for testclip. Enable the Overlay() line to show the testclip over the resulting video. That caught most of the frames with residual interlacing. It also screwed up a few frames without interlacing because something in the picture fell in the right frequency range. You might also try using a mask of where the comb artifacts are to limit the patching to just those areas.


[Attachment 52975 - Click to enlarge]

The video probably also needs decimation down to 24p.

Thanks! I would much prefer a slightly blurrier frame

I have a question. The video was originally SD (854x480). Should I de-bicubic down first before I do this?

And if I do masks, it would probably be in another video editor, though.

Also, when I use tdecimate(cycle=25) there are still a considerable amount of jumpy sections do to left over duplicates. Any ideas on how to fix this?

Also, just for future sake, do you know the technical reason why this happens during fades and scene changes? I probably wont understand what you say, but you can try lol.

Also (sorry lol), what about the frames that it dosen't detect automatically?

[johnmeyer]

Maybe johnmeyer was viewing the GoogleDrive playback rather than a downloaded version.

I just downloaded the video and played it on VLC, both with deinterlacing on, and with no deinterlacing.

I still see absolutely no artifacts (OK, with deinterlacing turned off, I do see some slight combing during the transitions).

I most certainly do not see any artifact resulting from resizing interlaced video without first deinterlacing. That step would not be needed for this video because I'm pretty certain it was never interlaced in the first place. More importantly, the artifacts produced by doing that improper resizing are not subtle, but are massive and overpowering.

Whatever the OP is seeing here is pretty subtle.

[embis2003]

Whatever the OP is seeing here is pretty subtle.

I disagree that its subtle. And its not just me seeing it. Everyone in this thread has seen it.

I'm pretty certain it was never interlaced in the first place

How is there combing if the source material was never interlaced?

Also, thanks for clarifying it wasn't a result of resizing before deinterlacing. But does that mean this thread wasn't either?

OK, with deinterlacing turned off, I do see some slight combing during the transitions

Wait... so if you have deinterlacing ON you don't see them???

[davexnet]

This footage must have been a 24 or 23.976 > 25 fps conversion. That why there's a dup frame every second.
The blended frames come before a scene change, quite common, but to really understand what happened you need the source file.

[DB83]

I'll add a few words since the OP was courteous in his response.


I checked the VEVO link. Maybe 854*480 is the original (just as with that other promo) but it is actually the only one that my capture software did not pick up. But it did pick up 640*360,960*540,1280*720,416*234,480*270 AND 1920*1080.


I did, out of interest, grab the 1080p version and, yes, it is 25fps but, strangely, the bitrate was slightly lower than that available above. And the final file is 162 mb


There is even a 1920*1080 mpeg with vbr naturally higher than the AVC. And the vid is over 500 mb.


My obvious conclusion is that it is VEVO (just as yt does) that does the re-scaling. It may even, in this case, have screwed up with the frame-rate. Maybe the label was instrumental in cropping from an original 4:3 'Master' or maybe even VEVO can also do that on-the-fly.


I realise that little, if any, of this helps you but this was really for my own curiousity. I even now wonder if that other promo is also available from this source.

[embis2003]

I'll add a few words since the OP was courteous in his response.

I have no hard feelings towards you.

I checked the VEVO link. Maybe 854*480 is the original (just as with that other promo) but it is actually the only one that my capture software did not pick up. But it did pick up 640*360,960*540,1280*720,416*234,480*270 AND 1920*1080.

Yeah, I'm pretty sure UMG sends out a master file that is 854x480 to all of their partners (youtube, vevo, etc) and vevo in this case made different conversions with a corresponding bitrate, similar to YouTube. However, unlike youtube, vevo automatically upscales SD sources as opposed to simply having a higher bitrate.

I did, out of interest, grab the 1080p version and, yes, it is 25fps but, strangely, the bitrate was slightly lower than that available above. And the final file is 162 mb

There is two 1080p versions you can grab. One with a bitrate of 4200 and one with 5200. I think you just grabbed the wrong one.

There is even a 1920*1080 mpeg with vbr naturally higher than the AVC. And the vid is over 500 mb.

Wait... what? On the vevo page?

My obvious conclusion is that it is VEVO (just as yt does) that does the re-scaling. It may even, in this case, have screwed up with the frame-rate.

I would think the frame rate isn't modified, however, the youtube version of this video (if you look at the nerd stats) suggest the frame rate is 24. Therefore, you may be right.

Maybe the label was instrumental in cropping from an original 4:3 'Master' or maybe even VEVO can also do that on-the-fly.

I doubt it was originally a 4:3 master because the video is higher resolution than the VOB file. If you compare the two, the VEVO version has signifcantly more detail, more than a simply higher bitrate could provide. Intrestingly, the vob file is letterboxed 1.85:1, but the vevo version has been cropped to 16:9 by cutting 37 pixels off on both sides. So, the master was most likley 1.85:1 just like the vobs.

I realise that little, if any, of this helps you but this was really for my own curiousity. I even now wonder if that other promo is also available from this source.

It is: http://embed.vevo.com/?video=USIV20400678

However, strangly, it dosen't have those artifacts like it though, suggesting that it came from a different master OR the original master file (before it was converted to AVC) was interlaced 16:9 video with a letterboxed 1.85:1 stream. Perhaps, the master file I had was cropped from that, resized to 16:9 (without being deinterlaced which caused those horrible artifacts.

[DB83]

Yes. The 4200k is the AVC. The 5200k is the mpeg2. There was no reference to another AVC to give a file size of 199mb. Yet those numbers are misleading since the mpg2 has an avg bitrate of 11mbps.


Might check that other vid later.


Thanks.

[embis2003]

Yes. The 4200k is the AVC. The 5200k is the mpeg2. There was no reference to another AVC to give a file size of 199mb. Yet those numbers are misleading since the mpg2 has an avg bitrate of 11mbps.


Might check that other vid later.


Thanks.

http://22-cf-168198817351468082649c6ba82cee92.vip1-dal1.dlvr1.net/ed3c085a-8a07-11ea-a...0_aac_128.m3u8 is the stream I downloaded to get the file above.

[jagabo]

I have a question. The video was originally SD (854x480). Should I de-bicubic down first before I do this?

Probably not.

Also, when I use tdecimate(cycle=25) there are still a considerable amount of jumpy sections do to left over duplicates. Any ideas on how to fix this?

Music videos like this are typically telecined from film then edited as video. Shots are slow down or sped up as interlaced video to match the music. So you can't get smooth motion of all shots at 23.976 fps. This video has the additional problem that it was partially inverse telecined and converted to 25 fps with an addition frame every 24 frames. If you had the original 30i video you could smart bob it to 60p. But you don't.

Also, just for future sake, do you know the technical reason why this happens during fades and scene changes?

My guess in this case is they had a mix of telecined film and 30i camcorder video. Then cut/pasted without consideration for the pulldown pattern and crossfaded between shots as 30i. So it's not possible to cleanly IVTC blended sections.

Also (sorry lol), what about the frames that it dosen't detect automatically?

I don't understand the question.

[embis2003]

I have a question. The video was originally SD (854x480). Should I de-bicubic down first before I do this?

Probably not.

Also, when I use tdecimate(cycle=25) there are still a considerable amount of jumpy sections do to left over duplicates. Any ideas on how to fix this?

Music videos like this are typically telecined from film then edited as video. Shots are slow down or sped up as interlaced video to match the music. So you can't get smooth motion of all shots at 23.976 fps. This video has the additional problem that it was partially inverse telecined and converted to 25 fps with an addition frame every 24 frames. If you had the original 30i video you could smart bob it to 60p. But you don't.

Also, just for future sake, do you know the technical reason why this happens during fades and scene changes?

My guess in this case is they had a mix of telecined film and 30i camcorder video. Then cut/pasted without consideration for the pulldown pattern and crossfaded between shots as 30i. So it's not possible to cleanly IVTC blended sections.

Also (sorry lol), what about the frames that it dosen't detect automatically?

I don't understand the question.

How can I run that script on a frame that it dosen't detect?

[jagabo]

How can I run that script on a frame that it dosen't detect?

You can try adjusting the testclip so that more combing is detected. Increase the vertical sharpning, resizing to a different vertical size, or with a different resizer.

You can also try playing with the downscale size of the patch video. Downscaling less will blur less but will leave a little more combing. Find a compromise you can live with.

For a short video like this it's probably easier to just replace the frames manually with ReplaceFramesSimple() -- included in the RemapFrames package. So instead of using ConditionFilter() use:

ReplaceFramesSimple(last, patch, Mappings="[7864 7867] [7869 7871] [7873 7876] 7880 7886 7888")

"[7864 7867]" means replace frames 7864 to 7867 inclusive. Just a number means replace that one frame. I'll leave it up to you to find all the frames that need replacing.

[DB83]

Yes. The 4200k is the AVC. The 5200k is the mpeg2. There was no reference to another AVC to give a file size of 199mb. Yet those numbers are misleading since the mpg2 has an avg bitrate of 11mbps.


Might check that other vid later.


Thanks.

http://22-cf-168198817351468082649c6ba82cee92.vip1-dal1.dlvr1.net/ed3c085a-8a07-11ea-a...0_aac_128.m3u8 is the stream I downloaded to get the file above.

There is a very slight variance between my 1080p AVC


This is the 5200K that I downloaded


http://22-fa-168198817351468082649c6ba82cee92.vip2-ams1.dlvr1.net/35593fc2-89ef-11ea-a...0_aac_128.m3u8


and despite the link stating h264 it is mpeg2 unless mediainfo has it wrong (yet they recc the others as AVC so that is unlikely)


None of this really matters though


I did check out the other promo now with this new link and all the versions available are mpeg2 AND 25fps. Weird.

[embis2003]

Yes. The 4200k is the AVC. The 5200k is the mpeg2. There was no reference to another AVC to give a file size of 199mb. Yet those numbers are misleading since the mpg2 has an avg bitrate of 11mbps.


Might check that other vid later.


Thanks.

http://22-cf-168198817351468082649c6ba82cee92.vip1-dal1.dlvr1.net/ed3c085a-8a07-11ea-a...0_aac_128.m3u8 is the stream I downloaded to get the file above.

There is a very slight variance between my 1080p AVC




This is the 5200K that I downloaded


http://22-fa-168198817351468082649c6ba82cee92.vip2-ams1.dlvr1.net/35593fc2-89ef-11ea-a...0_aac_128.m3u8


and despite the link stating h264 it is mpeg2 unless mediainfo has it wrong (yet they recc the others as AVC so that is unlikely)


None of this really matters though


I did check out the other promo now with this new link and all the versions available are mpeg2 AND 25fps. Weird.

??? im not sure whats going on here? can you send me this file???


[Attachment 52985 - Click to enlarge]

[DB83]

Sorry. 535 mb is a rather large file to upload on a slow connection. But the mediainfo report should show I am not making this up.

[hintlou]

Is it the vevo still existed?
I'm in China.I wanted to collect the HK vevo MVs,but I couldn't connceted to VEVO even in a HK route or USA route VPN.
It will turn to youtube when I click the links like "vevo.ly/xxx"
As I know,there are some ways can download high bitrate files in vevo.com.I get a file of this kind from others and I know the file is from vevo.com certainly.
But nobody would tell me the secret of download methods.

[embis2003]

Is it the vevo still existed?
I'm in China.I wanted to collect the HK vevo MVs,but I couldn't connceted to VEVO even in a HK route or USA route VPN.
It will turn to youtube when I click the links like "vevo.ly/xxx"
As I know,there are some ways can download high bitrate files in vevo.com.I get a file of this kind from others and I know the file is from vevo.com certainly.
But nobody would tell me the secret of download methods.


[Attachment 52987 - Click to enlarge]

[Attachment 52988 - Click to enlarge]

If there is a way to get a higher bitrate from VEVO, I would love to know, but I don't.

[DB83]

I might be able to throw some light on the mystery of the mpeg.


I use 3 different download programs two of which are the same program (Jaksta) but different versions and are configured differently although neither are set to convert after completion. When I first clicked on the link, both were running. One was 'recording' mpegs (same as it did with the other vid) and one 'downloaded' AVCs. The mpeg was not available on the download. Even so the only AVC it sees for download is 164 meg. The problem with both of these programs is that they insist on grabbing every single 'available' version (no 199mb AVC though) and even then grabs duplicates which could be an issue with the manifest. You can stop the process but since you do not see their qualities until completion you might miss a better one.


Now just for a 'giggle' I fired up the other program (TubeDigger). Now that also sees many versions but only downloads what you ask it to. Interesting tho is that TubeDigger does not see mpegs or AVC mp4 but sees AVC ts files. I grabbed the highest quality version which has an even slightly higher video bitrate than the one you have and media info even report variable frame rate. This file is 216 mb. I think I can manage an upload of this one just for you to see if it makes any difference.

[embis2003]

I might be able to throw some light on the mystery of the mpeg.


I use 3 different download programs two of which are the same program (Jaksta) but different versions and are configured differently although neither are set to convert after completion. When I first clicked on the link, both were running. One was 'recording' mpegs (same as it did with the other vid) and one 'downloaded' AVCs. The mpeg was not available on the download. Even so the only AVC it sees for download is 164 meg. The problem with both of these programs is that they insist on grabbing every single 'available' version (no 199mb AVC though) and even then grabs duplicates which could be an issue with the manifest. You can stop the process but since you do not see their qualities until completion you might miss a better one.


Now just for a 'giggle' I fired up the other program (TubeDigger). Now that also sees many versions but only downloads what you ask it to. Interesting tho is that TubeDigger does not see mpegs or AVC mp4 but sees AVC ts files. I grabbed the highest quality version which has an even slightly higher video bitrate than the one you have and media info even report variable frame rate. This file is 216 mb. I think I can manage an upload of this one just for you to see if it makes any difference.

I'd really appreciate that! Might as well get as a high of a bitrate as I can. Also, curious about the vfr if that will help decimation.

[DB83]

Here you go:


https://drive.google.com/file/d/1b5ywDny6I2pVcd4Y0ZAXe4giOCLCUaV8/view?usp=sharing

[hintlou]

It's not the same kind of what I said.Your TS file is as the same as the file ofembis2003's link.
It's easy to get this 4200 or 5200K vevo AVC TS.
The method to get higher bitrate file I showed is still a secret....

[jagabo]

Here you go:


https://drive.google.com/file/d/1b5ywDny6I2pVcd4Y0ZAXe4giOCLCUaV8/view?usp=sharing

That's exactly the same video, just in a TS container rather than MP4.