How do we identify that videos filtered by QTGMC?

[Sach3932]

Can someone help me with QTGMC
I followed the Andrew Swan and rrats YouTube channel to install Avisynth+ and required plugins of version 64bit.
I already installed Avisynth+ 64 bit with necessary plugins along with VirtualDub2 in my machine but as I said when I add the script and make the changes but no changes will shown in VirtualDub2 preview.

Wanted to 480p video make it better using QTGMC please help me out.

[Selur]

What does your script look like?

[Sach3932]

SetFilterMTMode("QTGMC", 2)
FFVideoSource("McCartney-MPEG2_1080-interlaced_sample.mkv")
ConvertToYV12(interlaced = true)
AssumeTFF()
QTGMC(preset="slow", EdiThreads=3, matchpreset="slow", matchpreset2="slow", fpsdivisor=2, sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=1.2)
Spline64Resize(1920,1080)
Prefetch(10)

BUt still not getting the proper result. PLease find the attached copies of the video which i am trying to test QTGMC.

[Selur]

Your before&after show different frames. The 'before' has combing artifacts the 'after' not, so QTGMC does seem to do its job.
You won't get rid of those ugly transmission and compression artifacts with QTGMC.
Would need a sample to give good advice on that, but SpotLess, RemoteDirtMC and similar or ai based stuff like BasicVSR++ (<- would require Vapoursynth and a fast NVIDA CPU) might help.

SetFilterMTMode("QTGMC", 2)
FFVideoSource("McCartney-MPEG2_1080-interlaced_sample.mkv")
ConvertToYV12(interlaced = true)
AssumeTFF()
org = last
QTGMC(preset="slow", EdiThreads=3, matchpreset="slow", matchpreset2="slow", fpsdivisor=2, sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=1.2)
Interleave(org.Subtitle(text="Original"), last.Subtitle(text="Filtered"))
Prefetch(10)

might help spot differences


Cu Selur

[Sach3932]

I don't understand what you are trying to say could you please elaborate a little bit.Also, you mean to say that QTGMC is working on my machine. If yes then why the video looks too dirty or ugly as compared to the original video?

[Selur]

The original will have the same artifacts, if you look at the same frames, which you did not.
QTGMC is a bob deinterlaced, so it takes each field and interpolates each field to a frame, with fpsdivisor=2 you tell it to throw way half of the interpolated frames.
If in your source, one field might be damaged and the other not (or not as much) and if you now throw away the frames that were interpolated from the more damaged frames the output might seem more damaged, but in truth it is not.
http://avisynth.nl/index.php/QTGMC
Try what happens if you remove fpsdivisor=2 and add SelectOdd() below the QTGMC line. (iirc. by default QTGMC fpsdivisor=2 is the same as using SelectEven() after QTGMC)


Cu Selur

[Sach3932]

Thank you so much QTGMC is working fine on my machine.
Could you please provide the necessary filter script to make 480p clip better using QTGMC.

[Selur]

a. From your screen shot your source isn't 480p, but might be 480i.
b. QTGMC is the wrong tool for the job of cleaning up a source with such artifacts. You will need to use a different filter for that.
=> if you share a snort sample of your source, folks here probably can give you a few suggestions. Screenshots do not really say much about noise, that often is temporal, not static.

[Sach3932]

Here is another sample short clip which contains 480p. The previous one which I am showing is 1080i with interlaced video.
Provide the necessary filter list to make SD videos of better quality.
Thank you for your time.

[Sach3932]

Also, I have read somewhere that they show some movies with QTGMC+AIE what it mean could you please elaborate?
How they restore old videos using QTGMC?

[Selur]

QTGMC + AIE usually stands for
1. deinterlacing with QTGMC (this is only necessary, makes sense on interlaced input)
2. upscaling using ai enhanced methods (+ often some ai based filtering to denoise)
(so it's QTGMC + artificial intelligence enhancement)
But this will not create miracles and requires adjustment of the filters for each source.
Some folks do this either in Vapoursynth or, by using QTGMC in Avisynth/Vapoursynth and then upscaling the result with, for example: Adobe or Topaz Labs tools.

Looking at your 'Interlaced Sample VHS.mp4':
a. it's progressive AVC (so already deinterlaced)
b. it's way too compressed to rescue many details.
c. it was resized before deinterlacing, while not taking the interlacing nature of the video back then into account.


If you tried to get rid of those wavy lines in the skirt, you would have to throw away even more of the details.
If that is your source, you got no real hope. If you got the original interlaced content or the source VHS, there is still hope.

How they restore old videos using QTGMC?

a. They to a proper capturing of the source into a lossless interlaced format.
b. They use QTGMC in Avisynth or Vapoursynth for deinterlacing.
c. They do additional filtering and color corrections using Avisynth, Vapoursynth or some NLE.
d. They use Vapoursynth or tools from Adobe or Topaz Labs.


Your second source is also messed up. There, interlaced content was improperly resized and then saved as progressive content, which destroyed the interlaced structure.

You will not get anything that really is good out of those sources, since unlike those QTGMC + AIE examples you have in mind your source material (assuming those clips are your sources) are way worse than what was used for QTGMC + AIE encodes.
So something like:

or with (really slow and sketchy face enhancement)

is probably the best you can archive if you properly learn to use Avisynth, Vapoursynth and some other tools. (I quickly created these using Vapoursynth and some ai filters.) Other scenes will look way worse.


=> your first goal should be to get better sources, your second should be to understand video filtering&co.
There is no simple to use and 'one-click'-tool which will get you something good.
The nearest to that is probably using Topaz Labs stuff for everything, but even then, if you don't combine tons of stuff, your output will probably look bad.

Cu Selur

[Sach3932]

Can you please guide me how do you do that in vapoursynth and which AI filter did you used to achieve this. So I can give a try.Could you please provide the vapoursynth script and ai filters.
Which one is the best Avisynth or Vapoursynth as per your thoughts please suggest

[Sach3932]

How to identify that sources are good or not.
To achieve best result do I need to make the video inetrlaced or QTGMC is only work with interlaced videos not with the progressive videos?

[Selur]

What GPU do you use? Unless you use a modern NVIDIA GPU the CodeFormer (which I used for the face enhancement) will not work (same for a bunch of other ai filters).

Which one is the best Avisynth or Vapoursynth as per your thoughts please suggest

That totally depends, both have their pro and cons.
I usually prefer Vapoursynth since it's based on Python, which allows adding Python based filters and gives a more consistent scripting experience in my opinion.

How to identify that sources are good or not.

Mainly experience.
You know different artifacts (https://www.avartifactatlas.com/) and understand that lost details can't be recovered.

To achieve best result do I need to make the video inetrlaced or QTGMC is only work with interlaced videos not with the progressive videos?

For best results, you properly capture the source into a lossless format. If the source is interlaced your capture should be interlaced after capturing, if it's progressive, your captured file should also be progressive.

QTGMC has two main purposes:
1. undo interlacing (not !telecine!)
2. repair light residual artifacts after deinterlacing
in these steps it also supports some denoising and sharpening, but often using separate filters for this is to be preferred.
QTGMC is not some magical restoration filter like you seem to believe.

Cu Selur

[Sach3932]

I have windows 11 i7 11th gen RTX 3060 6GB Graphics with 16GB RAM and 1tb SSD
Can you please guide me how to use vapoursynth.

I don't understand what is mean by capture progressive video in lossless and then use the QTGMC filters to make it better?

[Selur]

I don't understand what is mean by capture progressive video in lossless

Okay, then you first should read up on what it means that a video is progressive or interlaced.
Your graphic card might just work for SD content, but it will be really slow (probably 1fps or similar).

Easiest way to use Vapoursynth with ai tools is probably using Hybrid with some addons.
Since you just showed, that you do not even know the basics about video storage formats and processing, Hybrid would probably a bad idea, since it assumes that the user knows what he is doing, and you clearly do not.
=> Read up about video processing, video color formats, aspect ratio notations, (de-)interlacing, telecine, pull-down, pull-up and once you at least a basic understanding Hybrid might be worth a try. (There is no tutorial or manual for Hybrid, so if you don't know how video encoding guis work, you will probably be lost.)
Maybe even using something like chatgpt 3 might work to understand some basics.

Cu Selur

[Sach3932]

Can you please elaborate this For best results, you properly capture the source into a lossless format. If the source is interlaced your capture should be interlaced after capturing, if it's progressive, your captured file should also be progressive.

[Sach3932]

Could you please provide the script which you have used to make the video better which you shown me in screenshots.
I have installed the hybrid on my machine and also used that little bit. Could you please provide the basic video screnshot that how to use vapoursynth or Avisynth using hybrid so I can give a try on my machine

[Sach3932]

How do you know that Your graphic card might just work for SD content, but it will be really slow (probably 1fps or similar). Need to understand

[Sach3932]

How to use CodeFormer?

[Selur]

Could you please provide the script which you have used to make the video better which you shown me in screenshots.
I have installed the hybrid on my machine and also used that little bit. Could you please provide the basic video screnshot that how to use vapoursynth or Avisynth using hybrid so I can give a try on my machine

No. If you can't figure that out on your own, you definitely need to read up on video processing&co.

How do you know that Your graphic card might just work for SD content, but it will be really slow (probably 1fps or similar). Need to understand

ai filtering requires tons of resources, my typical Vapoursynth scripts with ai filters requires ~24GB or RAM and my 16GB of VRAM is maxed out and on my Geforce 4080 the process isn't fast, so on your 3060 with just 6GB VRAM you will have to tile the content in the ai filters which will slow things down even more. So my estimate is that you probabl will end up around 1fps.

How to use CodeFormer?

CodeFormer (https://github.com/sczhou/CodeFormer) can be used through Vapoursynth vs-codeformer (https://github.com/HolyWu/vs-codeformer), which is part of one of the addons for Hybrid.

---
btw. asking for 'best' results is considered stupid, if you can't give an objective measurement that allows folks to know what you consider 'best'.

Cu Selur

[Sach3932]

Can you please elaborate this For best results, you properly capture the source into a lossless format. If the source is interlaced your capture should be interlaced after capturing, if it's progressive, your captured file should also be progressive. What is mean by this??

[lollo]

Can you please elaborate this For best results, you properly capture the source into a lossless format. If the source is interlaced your capture should be interlaced after capturing, if it's progressive, your captured file should also be progressive. What is mean by this??

It means: For best results, you properly capture the source into a lossless format. If the source is interlaced your capture should be interlaced after capturing, if it's progressive, your captured file should also be progressive.

[Selur]

it means that you need to capture VHS:
a. into a lossless format (not highly compressed AVC)
b. if your source is interlaced, you need to capture into an interlaced format (not a progressive one)
c. if your source is progressive, you need to capture into a progressive format (not an interlaced one)
If you don't know what interlaced and progressive means, read: https://www.studiobinder.com/blog/interlaced-vs-progressive-scan/

=> so this was my last post in this thread since you clearly do not care to learn stuff you just aim to get stuff 'spoon-fed' and at least I don't have the time for that.

[Sach3932]

Ok got it but I want to ask that how to capture the source into lossless format or in which software do I need to capture the source?

[Sach3932]

@ Selur I really apologize for my stupid concerns. I just wanted to ask that could you please guide me what are things needs to be learn in way like first to learn about video interlaced or progressive after that QTGMC Filters, Avisynth+ and vapoursynth and hybrid is this correct flow to learn so I can follow this.

[Sach3932]

I am really appreciate to you that you helped me a lot to understand lots of the things just one last questions to I don't have VHS source videos, I have MP4 videos or 480 p format so those videos I want to make it better using QTGMC and ai filters that's my motive.
And again I really apologize to you��

[lollo]

I have MP4 videos or 480 p format so those videos I want to make it better using QTGMC and ai filters that's my motive.

Then you have to put some effort on your side, and not just asking for a baby-sitting step by step

[Sach3932]

I am using this repo to install BasicVSR++
https://github.com/HolyWu/vs-basicvsrpp
I am trying with the following commands.

pip install -U openmim - It runs properly

mim install "mmcv>=2.0.0" - Stuck on this command its saying following message.

C:\Users\SACHIN>mim install "mmcv>=2.0.0"
'mim' is not recognized as an internal or external command,
operable program or batch file.

Anyone have idea about it?