The old "war of opinion" - Capture 8mm tapes in DV or lossless?

[Marvolo]

Ok everyone.

Here's a new attempt: comparing DV-AVI and uncompressed YUF capture (NO HUFFYUF labeling this time).

As was wished, this time with outdoor scenery. I can clearly see differences between them.

Also, no re-encoding has been done this time. I opened the uncompressed capture file directly in virtual dub, set the input and output frame as Brad explained and then saved it as AVI. No conversion to Huffyuf as VirtualDub couldn't offer me any Huffyuf codec. A ton Blackmagic 8bit and 10bit codecs popped up and the old Intel Video, but no Huffyuf. That's why I skipped the conversion.

Just be aware of the interlacing (as I now learned seem to be normal for uncompressed AVI files).

[Brad]

Why bother ? All you gain is 50% in disk storage.

To upload a smaller, lossless sample here instead of uncompressed.

Possibly because vdub requires either a composite or s-video analogue signal. You are feeding it a digital signal.

No. VirtualDub doesn't know or care what the source is. It provides access to the driver settings, and when set to the proper input, VirtualDub will just capture whatever bits the driver feeds to it.

I've captured HDMI in VirtualDub using multiple cards that have both analog and digital inputs, including the Blackmagic Intensity Pro 4K.

It's possible that the original Intensity Pro isn't compatible with VDub, or it may just be a settings problem. I thought I had read people using it in VDub years ago, but I'm just operating off memory for that one.

[Marvolo]

To upload a smaller, lossless sample here instead of uncompressed.

I'm sorry, lol. But VirtualDub didn't offer any HuffYuf codec, or maybe I missed it?


[Attachment 71979 - Click to enlarge]

[DB83]

If you are using the original vdub then you have to physically install a huffyuv codec.

Vdub2 includes a ffmpeg-fork of huffyuv

[johnmeyer]

There is absolutely no reason to ever store video as uncompressed. Why?

1. Storage may be cheaper now, but it isn't free.

2. Even in 2023, not all systems can reliably stream uncompressed without hiccups.

3. I doubt you will be able to play uncompressed on any smart TV, phone, tablet, etc. You will be limited to a computer. The smart TV won't be able to handle the data rate, and the phone & tablet actually DO have storage limitations.

So, use lossless compression as soon as you can and then, once you've done your editing, convert it to the highest quality "delivery format" you can, but make sure that both the delivery codec as well as the high-quality settings you use can actually be played by your target playback devices. If this is entirely just for you and not for family members or clients, then this step will be relatively easy. I have to deliver to dozens and sometimes hundreds of clients, so delivery compatibility is extremely important.

My point is, you have to think about this project beyond just the initial capture operation.

[Sharc]

The "sample _uncompressed .avi" has crushed whites, loosing details in the snow. The histogram has gaps, (which is normally a minor problem as has been discussed elsewhere)
The "sample_DV.avi" has brights in the superwhites, which are however not clipped, means this can be easily adjusted in post. The histogram is smooth.

Both have the horizontal stripes which has been found before. Can also be fixed in post though, see post#57. So post processing may be required anyway.

[Marvolo]

@Johnmeyer

I don't plan to go the "uncompressed" route. But as you may have noticed, reading the last posts, at the moment I can only capture in uncompressed using the Blackmagic Media Express software.
After capturing, I will then convert it to HuffYuf and these will be my master / archival files.

From there, I can edit and render them out to MP4 or whatever format I want to view it in.

As for the comparison between DV-AVI and lossless, ya'll have to make do with the uncompressed version as this is the only one I have right now / at the moment.

----------------

Anyway, thanks, Sharc for the comparison. I kind of read it as if the DV-AVI was the winner here? Personally, I found the colours to be brighter and stronger in the lossless one compared to the DV-AVI one?

[lollo]

Here's a new attempt: comparing DV-AVI and uncompressed YUF capture ... I can clearly see differences between them.

I can see the difference as well, and conclude your efforts are worthy. If you had a shot with fine details in close-ups the difference would have been more evident.



Here the slider comparison https://imgsli.com/MTg4MDAy

To better compare I apllied to both the the same "light" AviSynth processing (just basic deinterlacing):

v1=FFmpegSource2(video_dir_1+video_1).converttoyuy2().assumeBFF().QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)

v2=AviSource(video_dir_2+video_2).converttoyuy2().assumeTFF().QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1).trim(1,0)

The "horizontal lines" are still evident (because the source), some recent thread in the forum considered this aspect and a potential mitigation.

edit: Sharc arrived first, I still think lossless is "better", whatever the levels

[Marvolo]

I can see the difference as well, and conclude your efforts are worthy.

Thanks. From what I've seen so far, I think so, too!

[Sharc]

Anyway, thanks, Sharc for the comparison. I kind of read it as if the DV-AVI was the winner here? Personally, I found the colours to be brighter and stronger in the lossless one compared to the DV-AVI one?

Yes, but colors can be aligned and adjusted as discussed earlier. Anyway, we compare file sizes which are orders of magnitude different, and the uncompressed version is of little practical use. I don't declare a winner, just pointing out some observations. One can always take criteria out of the equation and bias the "comparison".

[Marvolo]

Surely, my D8 recorder weighs in as well. When I re-captured all the tapes in DV-AVI with it (2023), while most of it was identical to the 2008 DV capture, I noticed that my cam seems to be brighter in gamma/brightness. Maybe that's playing into it now as well and probably the reason for the clipping in the snow?

I don't know which cam was used for the 2008 capture, but mine seems to make it a little brighter in general.

[Marvolo]

Here the slider comparison https://imgsli.com/MTg4MDAy

Very useful! That's how I usally compare them, too. Just in my NLE by muting one track and then the other and then see how the overall image changes. But it will change all at once, not gradually like here with the slider. I prefer the slider comparison now.

[johnmeyer]

Great comparison, lollo! I need to find out what service you are using to generate those "slider slides".

The increased saturation on the yellow is obvious, but it's tough to know if the more saturated version is better or truer to the original. Was the yellow really that intense compared to all the other colors in the live scene?

I was surprised that there was no apparent saturation difference in the blues.

I am not sure whether the blue snow highlights in the uncompressed version is what I'd want. That is one of the few things where I don't think the uncompressed version is better.

I zoomed in and looked for DCT compression artifacts in the DV capture. There is certainly a difference in some of the flat areas with little detail, but whatever I am seeing is awfully subtle.

I do see some slight difference in the diagonal lines around the eyes and mouth of the smiley face. Again, no clear winner, just differences.

People will always pick the TV image that is more saturated: it's an old retail showroom trick, so if you showed ten people both videos, actually playing, rather than single frame snapshots, the more saturated one is going to be picked.

However, if you edited one or the other in post so the saturation matched (which is easy), and then had a randomly-picked audience view them both, I wonder if there would be a consensus?

So, the uncompressed is, IMHO, a little better, but by only the slimmest of margins.

[poisondeathray]

You're getting a few more usable pixels on the right border for the DV version too

[Marvolo]

Johnmeyer

could the clearly noticible difference in saturation and brightness be soley due to the fact that this time the capture line was S-Video out --> Panasonic --> HDMI --> Intensity Pro Card?

When I capture the same tape in DV using the FireWire cable, I barely get a difference to the 2008 "professional"* capture. No increase in saturation like with the HDMI lossless route here.
I wonder where this massive difference comes from? Is this because of S-Video out?




* That said, I highly doubt now it was a "professional". From what I can see now, whatever they did was put my tapes in a random D8 camcorder, connected the cam via FireWire and hit record. I yield nearly identical results when I do the same now using Firewire and DV-capture.

@DB83
Thanks for the hint regarding VirtualDub2. Downloaded it and it does have the Huffyuf codec, like you said. That means, after capture in uncompressed, I will be able to convert it to Huffyuf and these will then be the archival files / master files.

[Sharc]

Both files of post#121 have bad glitches around frame#354. The damage is in the odd field only, the even field is mostly ok (or vice versa).
@Marvolo: Is the DV variant from 2008, or is it from 2023? Is the glitch also there in the 2008 'professional' version, or is it possibly due to the tape deterioration since 2008?

[lollo]

So, the uncompressed is, IMHO, a little better, but by only the slimmest of margins.

That's what we always said, john. The DV approach is easy and provides a quality result. Lossless approch (either with HDMI route or with classic recommended capture card workflow) has small advantage in term of quality, at the price of a much larger files, something that we also have to consider as Sharc properly noted (but both "master capture" will be probably compressed to h264 for distribution, so you only pay that penalty once).

Also consider that for noisy VHS captures the DV approach will suffer more than lossless approach than what we see here. And for NTSC you have to add the 4:1:1 color space subsampling instead of the PAL 4:2:0 (if this is really an issue it's still opinable according to someone).

In addition to what you pointed out in the comparison, I also find the edges of the small objects to be more accurate for the lossless capture, but the frame itself does not contain many elements to judge in a definitive way.

For the slider, I use VirtualDub lo load an AviSynth script (to align the captures, center, etc) and generate png images. I load them then to https://imgsli.com. Be careful that imgsli.com compresses the pictures, so the differences are softened a bit. In this specific case I also added a deinterlace operation (same to both) to do not have jagged edges confusing the comparison.

You can build a "slider comparator" in your web browser as well, if you need the details just let me know.

[Bwaak]

I don't see major differences in lollo's comparison aside of saturation and tint. Bluish snow in the HDMI capture does not have more detail. No other objects have more detail. I even think that the wooden siding on the house has more clearly visible edges on the planks on the DV version. HDMI version has more reddish noise. Then again, this is not my call.

[lollo]

You're getting a few more usable pixels on the right border for the DV version too

Excellent point.

The reason should be the Panasonic and the HDMI capture. With the classic lossless flow using the recommended capture cards the usable pixels should be the same as in the DV capture https://forum.videohelp.com/threads/410062-Which-capture-of-my-old-VHS-looks-the-best#post2694196

[DB83]

IIRC The only downside with the vdub2 version is the lack of control re bitrate etc. You might find that using a formal codec - v2.1.1 is the accepted one - still gives you the lossless quality but smaller files.

[Marvolo]

Both files of post#121 have bad glitches around frame#354. The damage is in the odd field only, the even field is mostly ok (or vice versa).
Is the DV variant from 2008, or is it from 2023? Is the glitch also there in the 2008 'professional' version, or is it possibly due to the tape deterioration since 2008?

Every DV file I have uploaded here so far for comparison with the lossless capture was from the 2008 (first) capture. So yes, that glitch you see. I saw it too and it was there in the 2008 file already and now again. So, seems to be a problem with the tape itself as it manifests at exactly the same place in different playback machines.

[johnmeyer]

could the clearly noticible difference in saturation and brightness be soley due to the fact that this time the capture line was S-Video out --> Panasonic --> HDMI --> Intensity Pro Card?

When I capture the same tape in DV using the FireWire cable, I barely get a difference to the 2008 "professional"* capture.

I don't know the answer. We cannot know the internal signal path inside of the Digital8 camcorder, so I don't know how the chroma and luma are separated before being encoded.

As for your HDMI capture chain, I have never been clear as to whether S-Video provides as much of an advantage for PAL transfers as it does for NTSC. For NTSC, the color carrier was shoehorned into an existing B&W carrier when color was added to the existing standard in 1953. Because of how it is done for NTSC composite video, it is almost impossible to cleanly extract the color, especially the red end of the spectrum, even with a comb filter. Thus, S-video makes a really big difference for NTSC.

Since both of your captures appear to be 4:2:2, I am not sure I know the reason for the significant saturation difference on the yellow smiley face when the other colors are the same as the DV capture.

[johnmeyer]

For the slider, I use VirtualDub lo load an AviSynth script (to align the captures, center, etc) and generate png images. I load them then to https://imgsli.com. Be careful that imgsli.com compresses the pictures, so the differences are softened a bit. In this specific case I also added a deinterlace operation (same to both) to do not have jagged edges confusing the comparison.

You can build a "slider comparator" in your web browser as well, if you need the details just let me know.

Thanks for that. I'll definitely try it out the next time I'm posting an "A/B" comparison.

[Marvolo]

Since both of your captures appear to be 4:2:2,

That's what bewilders me and I believe it can't be right. DV-AVI can never be 4:2:2, can it? I think there must be something wrong with these 2008 files in terms of data flags. They have always behaved differently to all the DV files that I myself have produced using my NLE over the years. I don't know what they used to capture them and to create these files, but I've frequently run into minor issues whenever I imported them into my NLE that I never had with my own DV files.

[Bwaak]

DV-AVI can never be 4:2:2, can it?

DVCPRO50 is 4:2:2, but it is a professional format.

[Sharc]

DV-AVI can never be 4:2:2, can it?

DVCPRO50 is 4:2:2, but it is a professional format.

Correct.

Chroma subsampling in DV is a smorgasbord:
DV25 (consumer, DVCAM) 480i 4:1:1, 576i 4:2:0
DV25 (D-7) 480i 4:1:1, 576i 4:1:1
DV50 (DVCPRO, D-9) 480i 4:2:2, 576i 4:2:2
DV100 (DVCPRO HD, D-12, Panasonic) 4:2:2

Reference: Charles Poynton's "Digital Video and HDTV Algorithms and Interfaces", chapter 39, DV compression.

Since both of your captures appear to be 4:2:2,

That's what bewilders me and I believe it can't be right. DV-AVI can never be 4:2:2, can it? I think there must be something wrong with these 2008 files in terms of data flags. They have always behaved differently to all the DV files that I myself have produced using my NLE over the years. I don't know what they used to capture them and to create these files, but I've frequently run into minor issues whenever I imported them into my NLE that I never had with my own DV files.

From MediaInfo it looks like the 2008 DV captures used a DVCPRO HD codec (4:2:2) from SONY, applied to the SD format (720x576i). Strangely though ffprobe reports the pixel format as yuv420p, chroma location top-left. MediaInfo and ffprobe reports are conflicting. Maybe a container vs stream conflict, I don't know.

[Marvolo]

From MediaInfo it looks like the 2008 DV captures used a DVCPRO HD codec (4:2:2) from SONY, applied to the SD format (720x576i). Strangely though ffprobe reports the pixel format as yuv420p, chroma location top-left. MediaInfo and ffprobe reports are conflicting. Maybe a container vs stream conflict, I don't know.

Keep in mind, though, that the DV files I uploaded on here are an re-exported version from the original 2008 DV files. To my understanding, no re-encoding has been done because my NLE offers what is called "Smart Rendering" for DV files. Meaning: if only hard cuts without VFX or other changes have been made, then it will render out without re-encoding into the same target format. I always make sure that "Smart-Rendering" is in place so that there is no lossy re-encoding.

That's what happened here.

Here is MediaInfo on the actual original 2008 file (without my NLE meddling with it):

[Attachment 71990 - Click to enlarge]

As soon as I use my NLE to re-export a fraction of it using the Smart-Rendering mechanism (no re-encoding in place), MediaInfo says this:


[Attachment 71991 - Click to enlarge]

No idea what this is all about.

[Marvolo]

DV-AVI can never be 4:2:2, can it?

DVCPRO50 is 4:2:2, but it is a professional format.

Of course, but I was thinking of regular consumer SD-DV, as you get when you record from a regular DV cam via FireWire. That's not 4:2:2. I wasn't even thinking of professional DV formats because as we all know this is not the case here for my files.

[Brad]

You're getting a few more usable pixels on the right border for the DV version too

Excellent point.

The reason should be the Panasonic and the HDMI capture. With the classic lossless flow using the recommended capture cards the usable pixels should be the same as in the DV capture https://forum.videohelp.com/threads/410062-Which-capture-of-my-old-VHS-looks-the-best#post2694196

Sorry, no.

From a classic FLP437 thread:

I noticed the D8 camcorder puts a solid black border on the left when it's outputting analog. The DV cap [WF3] has a smaller border on the left (like it's on the tape) because it does not run through the D8's DAC. Not really relevant but I was wondering what was causing it because the border is too sharp to be on the tape.

I had the same effect with my NTSC Digital8 cam's S-Video output. From old thread:

it's clear that the analog output always goes through a digitization stage regardless (chroma is always DNR'd, and hard borders are masked onto both sides).

[Brad]

Keep in mind, though, that the DV files I uploaded on here are an re-exported version from the original 2008 DV files. To my understanding, no re-encoding has been done because my NLE offers what is called "Smart Rendering" for DV files. Meaning: if only hard cuts without VFX or other changes have been made, then it will render out without re-encoding into the same target format. I always make sure that "Smart-Rendering" is in place so that there is no lossy re-encoding.

My VirtualDub "Direct Stream Copy" instructions were intended both for a Huffyuv source, and a DV source (plus any other codec in an AVI container). VDub shouldn't mess with the metadata like your NLE is.