Comparing two different 8mm (video8) capture attempts & different players

[Marvolo]

Start with a good player with integrated TBC

Does the Sony DCR-TRV730E has this by any chance? That's the one I bought from Ebay. It looked like a good camcorder to me because it also has analogue-in and DV-in, which seems to be rare.

[lollo]

Yes

[Alwyn]

Here are two files. The source video is a Video8 tape from 1994. One file is from a PAL DV file captured via firewire from my Sony TRV-110E Digital 8 camcorder; the other is from the same camcorder but captured using S-Video and my GV-USB2. Both were processed with VDub 2: deinterlaced Yadif double Frame rate, cropped and resized.

To my eye, there is no noticeable difference and therefore I would suggest sticking with the Firewire workflow for simplicity. I'll add that a DV workflow can utilise Scenalyzer, an amazing program for manipulating DV files which makes life so much easier than working with any of the other video formats.

The only consideration would be that the DV is from a PAL source, with 4:2:0 colour. NTSC DV is only 4:1:1 and so, allegedly, the DV quality will be less, although I have never seen an actual example of this lesser quality.

[Bwaak]

I would side with Alwyn. Unless you want to tinker, using built-in DV conversion gives you a predictable result in an industry-standard codec.

[Bwaak]

I would shift chroma one or two pixels up and a couple of pixels to the right.

I see a degradation doing that (also on the green logo). What am I missing?

https://imgsli.com/MTc4MzQ2


[Attachment 71002 - Click to enlarge]

That red bleeding can be removed with L=-4, but the rest of the image will be badly impacted (my conclusion was that is not a full chroma shift, but just a bleeding). I may be wrong.

Here a small sample of the raw capture, if you wish to experiment:


[Attachment 71003 - Click to enlarge]

Shifted chroma two lines up. I don't care much about the logo. I guess it may happen when elements added on top of the main picture may not exactly match, in this case I would choose the main picture.

[lollo]

Shifted chroma two lines up. I don't care much about the logo. I guess it may happen when elements added on top of the main picture may not exactly match, in this case I would choose the main picture.

Strangely there are part of the pictures not affected by chroma shift.

In general, if present, the shift in VHS PAL is distributed across the whole picture. Is is unlikely that there is a shif in the master video before the broadcaster applied the logo, but you never know.

Thanks for sharing your point of view

[MJTVID]

Here are two files. The source video is a Video8 tape from 1994. One file is from a PAL DV file captured via firewire from my Sony TRV-110E Digital 8 camcorder; the other is from the same camcorder but captured using S-Video and my GV-USB2. Both were processed with VDub 2: deinterlaced Yadif double Frame rate, cropped and resized.

To my eye, there is no noticeable difference and therefore I would suggest sticking with the Firewire workflow for simplicity. I'll add that a DV workflow can utilise Scenalyzer, an amazing program for manipulating DV files which makes life so much easier than working with any of the other video formats.

The only consideration would be that the DV is from a PAL source, with 4:2:0 colour. NTSC DV is only 4:1:1 and so, allegedly, the DV quality will be less, although I have never seen an actual example of this lesser quality.

Which file is DV and which is S-Video?

I compared screenshots overlaid and zoomed in Photoshop, and to my eyes video1 is easily better, slightly more detail with less pixelation. Video2 appeared sharper but that was just more pixelation with less actual detail. IMO

[poisondeathray]

Here are two files. The source video is a Video8 tape from 1994. One file is from a PAL DV file captured via firewire from my Sony TRV-110E Digital 8 camcorder; the other is from the same camcorder but captured using S-Video and my GV-USB2. Both were processed with VDub 2: deinterlaced Yadif double Frame rate, cropped and resized.

To my eye, there is no noticeable difference and therefore I would suggest sticking with the Firewire workflow for simplicity. I'll add that a DV workflow can utilise Scenalyzer, an amazing program for manipulating DV files which makes life so much easier than working with any of the other video formats.

The only consideration would be that the DV is from a PAL source, with 4:2:0 colour. NTSC DV is only 4:1:1 and so, allegedly, the DV quality will be less, although I have never seen an actual example of this lesser quality.

DV is usually a very solid choice... but beware that some programs do not handle interlaced 4:2:0 optimally - such as PAL DV

file 1.mp4 's chroma wasn't handled properly somewhere in the workflow - resulting in "chroma lag" or "chroma ghosting"

This is an apng comparing a cropped view of File 1 (top) and File 2 (bottom). It should animate in most browsers. Notice the color information for File 1 doesn't "fit" centered, it alternates left/right , creating a sort of flashing or flicker effect



It's likely from suboptimal 4:2:0 interlaced handling in vdub / vdub2 (the original vdub author was of the belief that interlaced 4:2:0 does not exist, so it handles interlaced 4:2:0 progressively. A potentially serious issue considering the sheer number of consumer cameras that shot interlaced 4:2:0 back in the 2000's-2010's , such as HDV, AVCHD, and the number of vdub users back in the day.) A workaround in vdub is to upsample to 4:2:2 first, or configure your DV decoder to send 4:2:2 . Or there are dozens of other workflows that do not have the interlaced 4:2:0 handling issue - avisynth, vapoursynth, ffmpeg , most NLE's (some have the problem, especially older non "pro" ones) etc...

[MJTVID]

Thanks for that. Most helpful and important to be aware of.

So video1 is the DV sample, which to my eyes has more detail with less pixelation/jaggedness. S-Video capture should be at least as good or better. I wonder if the S-Video capture went through a DAC and back to analog on a DCV-TRV*** series camera. I understand this occurs when TBC is enabled with these cameras. This would explain why the DV pixel structure looks better to me, aside from the motion color issue mentioned above.

[Alwyn]

A lot of water has flowed under the bridge since that post. I would deinterlace both with QTGMC now.

Here are snips of the two source clips. The LAGS clip was taken with AmarecTV, IIRC my GV-USB2. The levels weren't set specifically for that.

[poisondeathray]

Thanks for that. Most helpful and important to be aware of.

So video1 is the DV sample, which to my eyes has more detail with less pixelation/jaggedness. S-Video capture should be at least as good or better. I wonder if the S-Video capture went through a DAC and back to analog on a DCV-TRV*** series camera. I understand this occurs when TBC is enabled with these cameras. This would explain why the DV pixel structure looks better to me, aside from the motion color issue mentioned above.

The motion color artifacts are from mishandling of interlaced 4:2:0 ; not a criticism of the PAL DV format itself

I'd be careful about assessing any types of "artifacts", because those 2 videos are lossy re-encodes using AVC - which has inloop deblocking

Both videos use yadif deinterlacing, which contributes to the deinterlacing jaggy artifacts .

The 2nd video has capture issues as well - drop 640,641 , reverse motion 658,659

A lot of water has flowed under the bridge since that post. I would deinterlace both with QTGMC now.

Yes, you see the light

But many people still use vdub to deinterlace and are unaware of the issue. Another common one is DVD. Think of all the botched interlaced DVD processing jobs in the past and now... ugggh

[Sharc]

Yes, you see the light

But many people still use vdub to deinterlace and are unaware of the issue. Another common one is DVD. Think of all the botched interlaced DVD processing jobs in the past and now... ugggh

Here a script which visualizes flawed 4:2:0 interlaced/deinterlace handling. The chroma (U,V) advances with every second frame only.

LWLibavVideoSource("File 1.mp4",cache=false)
trim(470,520) #for demo
AssumeTFF()
Tweak(sat=2.0) #boosts the U,V visibility
StackHorizontal(last, Stackvertical(UtoY().subtitle("U"),VToY().subtitle("V")))
assumefps(2) #for demo

[MJTVID]

Thank you for the clarifications. S-Video to GV-USB2 has more detail and much better color depth than DV in these samples.

[clashradio]

You can adjust the degree of sharpening in QTGMC with the sharpness parameter as well.

Correct. I was just comparing the outputs without acting on sharpness parameters.
In general I use something like:

QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)

or some variant of it, based on some ideas I used from a discussion with Skiller.

If I understand correctly for the best deinterlace option, with no noise and sharpness reduction, use the above code? I also use Neat Video so I'd to use that for noise & sharpening. Or should I not even include "NoiseTR" and "sharpness"?

[lollo]

If I understand correctly for the best deinterlace option, with no noise and sharpness reduction, use the above code? I also use Neat Video so I'd to use that for noise & sharpening. Or should I not even include "NoiseTR" and "sharpness"?

There is not a best deinterlace option, the results strictly depends on the source, and the parameters of the filter should be tuned accordingly.

The given example aims to reduce the denoise and sharpening performed by QTGMC, a not easy task because these operation are part of the processing to obtain that nice outcome that we all know. Then a later denoise and sharpening filtering can be applied, hoping to do not over-process the whole, and do not introduce plastic look, halos and other defects.

Concerning denoise, for VHS/S-VHS captures an AviSynth / VapourSynth filtering is more performing than NeatVideo.

[clashradio]

If I understand correctly for the best deinterlace option, with no noise and sharpness reduction, use the above code? I also use Neat Video so I'd to use that for noise & sharpening. Or should I not even include "NoiseTR" and "sharpness"?

There is not a best deinterlace option, the results strictly depends on the source, and the parameters of the filter should be tuned accordingly.

The given example aims to reduce the denoise and sharpening performed by QTGMC, a not easy task because these operation are part of the processing to obtain that nice outcome that we all know. Then a later denoise and sharpening filtering can be applied, hoping to do not over-process the whole, and do not introduce plastic look, halos and other defects.

Concerning denoise, for VHS/S-VHS captures an AviSynth / VapourSynth filtering is more performing than NeatVideo.

I'm still confused about (preset="slow") vs fast. I thought slow does a better job at deinterlacing, but also applies noise reduction, where fast doesn't have any NR but has trouble with moving objects?

[lollo]

I never use "fast" preset except for some experiment where I need a quick deinterlace of a source. It does not disable completely noise and sharpening either, is just less accurate at its main goal.

In any case, always experiment with your own material, do not trust any "general" suggestion.

[rgr]

It's likely from suboptimal 4:2:0 interlaced handling in vdub / vdub2 (the original vdub author was of the belief that interlaced 4:2:0 does not exist, so it handles interlaced 4:2:0 progressively.

VDub treats every video as progressive by default. If you want to deinterlace, you need to use a filter.

[rgr]

Concerning denoise, for VHS/S-VHS captures an AviSynth / VapourSynth filtering is more performing than NeatVideo.

I will never agree with that. Sometimes I use AviSynth filters as a pre-filter, but the final effect is always best with Neat.

[rgr]

I'm still confused about (preset="slow") vs fast. I thought slow does a better job at deinterlacing, but also applies noise reduction, where fast doesn't have any NR but has trouble with moving objects?

https://macilatthefront.blogspot.com/2021/03/lets-test-qtgmc-settings.html

[poisondeathray]

It's likely from suboptimal 4:2:0 interlaced handling in vdub / vdub2 (the original vdub author was of the belief that interlaced 4:2:0 does not exist, so it handles interlaced 4:2:0 progressively.

VDub treats every video as progressive by default. If you want to deinterlace, you need to use a filter.

That video WAS deinterlaced. You probably didn't read the post carefully...

Here are two files. The source video is a Video8 tape from 1994. One file is from a PAL DV file captured via firewire from my Sony TRV-110E Digital 8 camcorder; the other is from the same camcorder but captured using S-Video and my GV-USB2. Both were processed with VDub 2: deinterlaced Yadif double Frame rate, cropped and resized.

[lollo]

Concerning denoise, for VHS/S-VHS captures an AviSynth / VapourSynth filtering is more performing than NeatVideo.

I will never agree with that. Sometimes I use AviSynth filters as a pre-filter, but the final effect is always best with Neat.

That's your problem

There was once a troll, with the same theory, but based on much better material than VHS/S-VHS source, which is an advantage. Master poisodeathray showed the facts: https://forum.videohelp.com/threads/403073-Why-is-Neat-video-the-best-video-denoiser/page11

[rgr]

It's likely from suboptimal 4:2:0 interlaced handling in vdub / vdub2 (the original vdub author was of the belief that interlaced 4:2:0 does not exist, so it handles interlaced 4:2:0 progressively.

VDub treats every video as progressive by default. If you want to deinterlace, you need to use a filter.

That video WAS deinterlaced. You probably didn't read the post carefully...

Here are two files. The source video is a Video8 tape from 1994. One file is from a PAL DV file captured via firewire from my Sony TRV-110E Digital 8 camcorder; the other is from the same camcorder but captured using S-Video and my GV-USB2. Both were processed with VDub 2: deinterlaced Yadif double Frame rate, cropped and resized.

So what's the problem that VDub treats progressive video as progressive?

[poisondeathray]

So what's the problem that VDub treats progressive video as progressive?

Progressive video is not the problem. And why would you deinterlace progressive video in the first place ???

[rgr]

Concerning denoise, for VHS/S-VHS captures an AviSynth / VapourSynth filtering is more performing than NeatVideo.

I will never agree with that. Sometimes I use AviSynth filters as a pre-filter, but the final effect is always best with Neat.

That's your problem

There was once a troll, with the same theory, but based on much better material than VHS/S-VHS source, which is an advantage. Master poisodeathray showed the facts: https://forum.videohelp.com/threads/403073-Why-is-Neat-video-the-best-video-denoiser/page11

Everyone can find their own truth

I've tested various denoisers on many videos and (I think) it always ended up with Neat doing the final denoising. Other denoisers could do it better, unfortunately at the cost of details or plastic look. But they are useful as pre-filters, because Neat also has its limitations.
And it has a decent GUI.
Only the sharpener is of average quality, but sufficient.

[lollo]

I've tested various denoisers on many videos and (I think) it always ended up with Neat doing the final denoising.

Same here, with analog captured sources. And find the opposite. As you said, each to his own!