Upscaling 480i VHS for modern 4K TVs?
 

With the technical knowledge now in 2022, what is the best way (hardware or software, understanding that hardware is the preferred method) to upscale a 480i video (VHS capture), so it looks as good as it possibly can when it is being viewed on modern 4k LED TV's? It is my understanding that VHS captured videos should be deinterlaced prior to upscaling.

It should be known that the videos will be on USB drives and plugged into a TV directly.
I know a preferred method is to leave the video interlaced and burned onto DVD but I prefer flash drives for the main reasons of scratched or broken discs and eventual incompatibility.

There is no need to go all the way to 4k from SD, Anything after HD is a matter of line doubling, deinterlace and upscale to 1440x1080, Your 4K TV will just have to double the lines, no heavy computation therefore no more quality loss.

With that said, what is the best way to go about it? I appreciate the response and the information. I have read that it is often recommended not to upscale above 720p, haven't heard reasons why much though.

It's better not to upscale at all and if you do have to upscale for whatever reason, the only resolution to avoid is 720p, It was a pro mobile camera resolution for sports and news gathering back in the day where progressive frames is needed (30p) for fast moving objects because 1080p was not possible yet, It became one of the standards for over the air broadcasting and quite few native 720p panels and consumer camcorders were made. Not anymore, it is as dead as DV, no 720p panels are being made anymore, Youtube dropped support for 720p and consider it SD now, your HD or 4K TV or monitor will have to re-upscale 720p files to HD or 4K on the fly anyway, so it's a middle man that is no longer needed. So no, don't make that mistake.

step 1, deinterlace if needed:

• if your video is not interlaced, but telecined PAL (the fields are from the same moment in time) you do not deinterlace
• if your video is not interlaced, but telecined NTSC you do not deinterlace, but apply IVTC with http://avisynth.nl/index.php/TIVTC or http://avisynth.nl/index.php/Decomb
• if your video is interlaced you deinterlace with QTGMC; QTGMC does denoise and sharpen by default, you may want to play with its parameters

step 2, change color matrix:

• when uscaling to HD the color matrix must be changed from standard 601 to 709

step 3, change resolution [and add DAR 4:3 flag because captured VHS files do not have it]

• use nnedi3_rpow2 to upscale by a power of 2 and resize to 1440x1080 (HD resolution)

step4, optional to build a 1920x1080 frame:

• add 240 black borders on left and right side [and change DAR to 16:9]

A PAL example:
step 5, optional to encode to h.264:

I use Topaz Video Enhance AI. It uses artificial intelligence to upscale the video. It works very well. There are several different AI models to try. It has models that work with interlaced as well as progressive video sources. I recommend you clean up the audio with Goldweave and do any color, contrast and other filtering before hand. I use Davinci Resolve Studio but you can use any program you are familiar with. I then render the video uncompressed then run it through Video Enhance AI. I am using Windows 11 a Mitsubishi HS HD2000U Svideo to a Diamond VC500 one touch capturing the video and My onboard Realtek for audio with Virtualdub64 using Utvideo 4:2:2 codec.

Very helpful! I haven't been doing this for whatever reason, will advise.

I use LanczosResize() or Spline36Resize() in these instances; what makes you prefer nnedi3? I'm interested in hearing your thoughts.

We can render entire Avisynth scripts outside of Virtualdub or Hybrid...? :eek:

That's terrible advice. Topaz isn't AI whatsoever. all marketing BS, and it creates lots of artifacts. The interlaced mode is beyond awful. You're essentially sticking video in a meat grinder.

Much longer discussion here: https://forum.videohelp.com/showthread.php?t=399360

Where/when did you get that VC500? Far too many reports of bad AGC in recent years.

- Most USB cards don't also external audio, it won't lock, desync.
- Onboard Realtek are very hit-or-miss, tinny, scratchy audio. Which specific board is it? I've seen a few good onboard Realtek, but mostly bad.

Topaz does a great job I really see no artifacts. Artifacts will be created if you use the wrong ai model.

I have read the agc complaints but my 7nit does not seem to have issues.

Could you post some samples? I would assume Topaz would have a hell of a time predicting noisy VHS, HD and film are good candidates for Topaz, But sooner or later it will be mainstream if it proves itself.

Topaz with aggressive settings creates interesting but unnatural results with VHS sources.

With less aggressive settings, it just sharpens the video a little, not worth the effort.

Maybe someday it will be useful for VHS upscaling. :praying:

So don't upscale, just use QTMC to deinterlace to 60p, and use the "grain" tuning in Handbrake to prevent macroblocking.

I had bad luck with Topaz software myself. Given it is a nice and simple process to get a deinterlaced video, it was not worth the quality loss I saw personally every time I tried it.

Has anyone ever tried out a Framemeister unit? I have seen some videos of it recently and show some promise of being able to deinterlace and upscale.

There is nothing that Topaz AI can do that cannot be beaten by AviSynth / VapourSynth (quote lollo2 and Selur)

nnedi3 features (slightly) better upascale results, but most important sharpen less. I generally do the sharpening after the upscale (because it is weaker than the other way), so I prefer the more "neutral" upscale.

I generally use a command line to have full control of the x264 parameters; however Hybrid is an excellent GUI, so you can easily use it.

The difference between traditional upscaled and ai upscaled is they sythesize pixels that are not there. This can create artifacts. I usually clean up as much as I can in low res then feed that to Topaz. My Mitsubishi HS HD200 has TBC and filters so it outputs a significantly cleaner capture. I can deinterlace before in resolve or avsynth. Tender to non interlaced,no compressed avi or other format then feed that to Topaz. I will upload samples.

I also find nnedi3 a decent middle ground. It's a more sophisticated upscaler than Lanczos and spline methods, using a neural network to try to connect lines/edges which helps reduce the jaggedness you tend to get on lines with the simpler methods. It won't match the best an AI upscaler like topaz could accomplish on a good source with the right models/tuning, but doesn't risk giving the weird artifacts like the fancier AI ones can sometimes do. It's also what QTGMC uses internally to scale stuff.

There are probably cases where the simpler methods could end up working better though if the starting footage is very bad.

In this experiment i feed Topaz with an AviSynth filtered source (a low res vhs capture), and compare versus the same AviSynth filtered source upscaled with nnedi3: https://www.youtube.com/watch?v=LW6XWp9e5ws

In this experiment I leave Topaz doing the full filtering and compare versus the Avisynth flow: https://www.youtube.com/watch?v=vmHJyF_XxT0

In both cases AviSynth is better.

Additional comparison (Topaz oversharpen): https://imgsli.com/MTAzNTg4

Some discussions we had in the past:
https://forum.videohelp.com/threads/...-AI-discussion (go through all pages to find nice samples)
https://forum.videohelp.com/threads/...Topaz-products
https://forum.doom9.org/showthread.php?t=183124

A common agreement we reached, was that the best approach is trying to use both AviSynth/VapourSynth and Topaz VEAI to obtain the best results. Topaz does not help in most of the cases. However, is worth a try for a quick a decent result .

Edit: we are talking about upscaling on VHS captures here!

Scalers for video games, like the Framemeister, have opposite goals compared to standard video content. For example, nearest-neighbour scaling can look excellent with 2D pixel art games. It looks horrendous with video.

On the flip side, if we want to run QTGMC to deinterlace, we may be willing to wait and let it run at 5 frames per second. Gamers want deinterlacing that gets frames through their entire display chain at a latency of less than half a frame, and are willing to accept ugly results for this benefit.

If you want to look at outboard scaling, the target demo we match with would be closer to home theater enthusiasts.
https://www.avsforum.com/forums/video-processors.37/