Should I upscale SD to HD during capture?

[powervideo]

I'm trying to develop a workflow before I transfer several hundred archived analogue tapes. The formats include low-band U-Matic, U-matic SP, VHS, S-VHS, Betacam SP and Digital Betacam. I have a good VO-9850 U-Matic VTR and two SDI-equipped Digi Beta playback machines. I plan to capture into my Blackmagic
Hyperdeck HD as ProresHQ via SDI.

I'm trying to weigh up whether I should get a Blackmagic UpDownCross SDI converter
https://www.videocraft.com.au/blackm...updowncross-hd

and put this in the chain in real-time to uprez the SD source material to HD, or if I should just capture the SD material as faithfully and non-destructive as possible, then do the up-scaling in post via software. There's so much material to transfer I don't want to make a mistake and find I've chosen the wrong path.

For example, while AI restoration and upscaling apps like Topaz are great, they're not perfect. Maybe it's better to digitise as SD and see what develops in AI archive restoration as it keeps developing, rather than do it now with a hardware solution during capture? Cause once it's done, it's done.

[lordsmurf]

No, never, you lose more quality with HD capture of SD material. The deinterlace is always rough and inferior, as is the upscaling.

There is nothing "AI" about Topaz software, nor is it really quality upscaling.

Blackmagic is a really mediocre company, in terms of products. Like Canopus in the late 90s and early 00s, BM is more about marketing than actual video quality.

The BM device you show is really more for news-type "live upconverting" where no time is available for proper upscaling. Sometimes obviously also used where quality is not cared about.

You'll never make a mistake by faithfully transferring the native data on the tape, using TBCs, retain proper size, interlace, and framerate. What is a mistake, what you will regret, is lossy converting it to the whizbang "best" method of the day. Because that day will pass. I've been doing this for 30 years now, and have seen this multiple times now.

Yes, once it (the capture) is done, it's done.
... unless you redo it, because you just f'd it up that bad -- and that also happens frequently, still, and for the past decade, correcting the mistake of the previous decade.

You are correct to worry.

[Aya_Rei]

Others can chime in if I'm wrong in a few areas, but it strongly is best to just capture the files in SD, 720x480, interlaced 29.97 FPS and at 4:2:2 colorspace. Making the raw file capture as true to the original tape as possible. Upscaling and deinterlacing through hardware isn't a good idea as they can be damaging to the footage, ie they don't a good job and make the final result worse. But also doing it through hardware isn't a good idea as upscaling and the like isn't being true to the original tape.

If you need to upscale, deinterlace and do script restoration, it is strongly encouraged to do that in software as it'll be way more flexible than hardware. Doing that stuff can be considered a part 2 to digitizing VHS, Video8, Beta, etc tapes (With capturing them as raw huffyuv avi files in VirtualDub being part 1). If you mess up with Avisynth for example and you'd want to try again with a tweaked script, then instead of converting the tape all over again. You can reuse your raw avi capture file over and over.

Just keep sure to backup those raw files, I do that to a portable SSD, along with backing up the final distribution mp4 copies to a separate 2TB external SSD

[aramkolt]

I understand the appeal of saving the time in not having to deinlterlace later. I thought I bought a decently specced PC, but it can only do QTGMC deinterlacing/scaling probably at best in real time, therefore every hour of content takes an hour to process. Good news is that most deinterlacers can run in batches, meaning they can run an automated list of files while you leave the computer completely unattended. Others have said they have better systems that can run at 450% of the realtime footage length, meaning a 2 hour video should complete in something like 30 minutes or less.

But, back to your question, you are the only one that can determine how likely you'd be to go back to the originals and reconvert later if better deinterlacing comes along and also determine if the bump in quality of using software over hardware scalers is worth the time it adds to your workflow. As long as the tapes themselves are in good condition and you aren't seeing them fall apart after one playback, you could capture them all via the upscaling method and go back and re-capture any of the most important content as interlaced and unscaled after you've skimmed the original capture's output. That'll also allow you to see if any of the tapes messed up during the first capture from a visual perspective.

The other option that isn't talked about much is capturing the content both ways simultaneously. Pretty much all SDI boxes have a passthrough option that you could then connect to something that captures lossless interlaced - I think ProResHQ is also considered visually lossless compression.

If it was me and I wanted to capture both ways, I'd go with something like a pair of AJA KiPro as the capture boxes - It'll capture SDI as 480i and has the option to input analog audio to mux to the SDI video signal (not sure if the digibeta puts audio into the SDI stream - but if it doesn't, you're covered there). You could then use that to capture from the passthrough of whichever hardware upscaler/deinterlacer you are using and then also attach a separate KiPro to the upscaled output of the Up/Down/Cross converter still set to capture ProResHQ in up to 1080P. You can tell the KiPro to pillarbox the widescreen content to preserve the aspect ratio, but may lose some horizontal resolution if you do it that way.

The KiPro can also scale as it records all by itself if you set it to - so it can scale 480i and record it as 480P, 720P, or 1080P to the drive all as it saves it as ProResHQ. I haven't played around to see if the progressive output it gives is at double the frame rate or not. I do plan to do a comparison of just about every hardware scaler out there and compare it to software options once I get my testing method established since this isn't something I want to do over and over given the number of devices I have to test. My theory is that there probably are hardware scalers that can give 80-90% of the quality of QTGMC can do and the question then becomes if for that last bit of quality if it's worth adding a bunch of steps or not - which would very much depend on the content. If it's a home video that is kind of blurry to begin with and maybe two other people will watch it one time, it may not be worth it. If it's some rare historical footage that you plan to share on the internet, I'd go for lossless capture and possibly the simultaneous upscaled capture.

The one quirk I've found with the limited testing I've done with the KiPro is that (at least with the analog audio inputs on it) it stores each of the L and R channels as a separate mono audio track. That could be in order to cram the highest bitrate/sampling rate into each maybe. Downside to that is that you'll later have to use software to create a stereo track if your source was stereo. As it is captured, you can only select one audio track to listen to at a time, or if you are then compressing, most programs such Handbrake, there is no option to take two separate mono tracks and turn them into one stereo track during the compression. Doing so without an intermediate step to make the two mono tracks into a stereo one essentially deletes the second audio track, and even if it did preserve it, you'd still only be able to hear one audio channel at a time. Hybrid might have a way to do that with the right script, but it isn't something that looks easy to do without some scripting knowledge.

You can create stereo audio from two mono tracks with FFMpeg and leave the video untouched as a separate step. I made a post about it here, though the whole automated script thing I came up with is if you are doing this on a Mac, but the actual commands should be similar on Windows: https://www.reddit.com/r/ffmpeg/comm...ono_tracks_on/

[powervideo]

Many thanks for all the detailed replies. I appreciate you putting the time into it. I think I'll capture in native SD interlace.

I do like recording directly into the SDI Hyperdeck. It's small, captures up to 1080 ProRes422HQ and even has a small LED monitor and VU on the front panel. The simultaneous approach is also attractive but I'm thinking I should just stick to one method of transfer; the best quality method. I'm sure up-conversion quality will incrementally increase in the coming years. I

These digitised reels will be broken up to put on our Pond5, Adobe, Shutterstock libraries, so It may not be too arduous deinterlacing in post if I don'y have to deintelace the whole reel. I'm going to start with the Betacams. I have two SDI decks, the Sony J30 and the SX DNW-A28. I've read that the SX machine is better with dropouts than the J30. It doesn't do AFM audio through; only linear. The J30 does both. I'll do some tests and post my thoughts here. Thanks again.