09-23-2021, 09:17 PM
|
|
Free Member
|
|
Join Date: Feb 2018
Location: Houston, Texas
Posts: 339
Thanked 64 Times in 57 Posts
|
|
I get it that SD video is captured at 480i. But I've got a project in mind where I'd like to make some downloadable YouTube videos using some footage which was taped aboard a steaming battleship in the 1980s. Obviously it's not possible to do a re-shoot! I'd like to edit in some still photos at points in the video when I want to show something in higher resolution. I'd like to make the final video 720p, or possibly 1080p if someone wants a Blu-ray copy. So, if I'm going to be resizing and enlarging SD video from a VHS source (lossless captures with an AIW card), what's the way to make it look as good as possible?
Got a render running from the capture right now; might be able to upload a few clips after it finishes.
|
Someday, 12:01 PM
|
|
Ads / Sponsors
|
|
Join Date: ∞
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
|
|
|
09-24-2021, 12:40 AM
|
|
Free Member
|
|
Join Date: Dec 2015
Location: USA
Posts: 3,530
Thanked 601 Times in 552 Posts
|
|
In today's standards 720p doesn't make any sense anymore, There is no native 720p displays left and it is not a multiple integer of neither 480 nor 1080, Therefore when you resize to 720p you get the first resize in software from 480 to 720 and the second resize during viewing from 720 to 1080/4K inside the TV or media player, since each resize is lossy you get double loss there, on top of that Youtube algorithm classifies 720p as SD and applies harsh compression and limits the frame rate to 30fps.
What you want if you must resize is, first de-interlace with QTGMC, crop the edges down to the active video area while respecting the ratio 704:480 for NTSC, then resize to 1440x1080 square pixel, and upload to youtube, That's exactly what I do for youtube.
A good reason why you resize to 1080 exactly and not more is that 1080 is a multiple integer of every future resolution, it will be just a matter of line doubling of future displays.
https://www.youtube.com/@Capturing-Memories/videos
|
The following users thank latreche34 for this useful post:
ehbowen (09-24-2021),
ThumperStrauss (02-01-2022)
|
09-24-2021, 04:55 AM
|
|
Invalid Email / Banned / Spammer
|
|
Join Date: Oct 2020
Location: London - UK
Posts: 568
Thanked 88 Times in 76 Posts
|
|
This won't be popular, but we're starting to dial in some good results with Topaz Video Enchance.
We've recently done a set of cautious 'smart' upscales and had some very good results, the deblocking and NR beats anything I've now seen with any other method including NeatVideo/Denoiser III and the usual AVISynth suspects, it doesn't take hours to do it either.
It's taken a bit of time to dial in some settings, but we're starting to use it more-and-more, we've done some full HD upscales, and whilst they don't look like HD footage they're more watchable on modern devices we're finding.
I know it'll be an unpopular opinion, but we're finding some very interesting results with it if you're cautious, feed it good footage, and experiment with the settings.
If you can post up a small sample (10 seconds or so) I'll quickly run it through our settings.
|
The following users thank RobustReviews for this useful post:
ThumperStrauss (02-01-2022)
|
09-24-2021, 06:09 AM
|
|
Free Member
|
|
Join Date: Feb 2018
Location: Houston, Texas
Posts: 339
Thanked 64 Times in 57 Posts
|
|
Quote:
Originally Posted by RobustReviews
If you can post up a small sample (10 seconds or so) I'll quickly run it through our settings.
|
I'd be willing to look at it, but ten seconds of lossless would exceed the max upload size of this forum. What format and compression do you suggest?
|
09-24-2021, 06:25 AM
|
|
Invalid Email / Banned / Spammer
|
|
Join Date: Oct 2020
Location: London - UK
Posts: 568
Thanked 88 Times in 76 Posts
|
|
Quote:
Originally Posted by ehbowen
I'd be willing to look at it, but ten seconds of lossless would exceed the max upload size of this forum. What format and compression do you suggest?
|
Let's try an experiment, send a few seconds of mp4, it's just an experiment.
|
09-24-2021, 07:00 AM
|
|
Invalid Email / Banned / Spammer
|
|
Join Date: Oct 2020
Location: London - UK
Posts: 568
Thanked 88 Times in 76 Posts
|
|
Quote:
Originally Posted by lollo2
|
Possibly, I'm often trying to shift several TBs of data in and out a day though.
I'm sure that many of these things can be replicated, however I work in quite a different market and can only report what I find.
I will have a flick through these when I get a moment, however without opening them I can feel pages and pages of debate, pages of AVISynth script etc which are great and I'm sure very, very good - however when you're working on about $6US/hour quite often for footage they're not practial.
But I don't want to start a debate there, I have my business model, others have theirs, those with time or restoring their own footage will have their own thoughts and opinions and rightly so. It's just interesting to experiment and let people draw their own conclusions on cost/benefit analysis.
You've also used 'Arthemis' in the comparison of yours that I opened, it's very poor with VHS footage.
For the avoidance of all doubt: This is just something I've been experimenting with, I know from previous experience that this could well be misrepresented as 'RR hates AVISyth, he destroys customers footage etc' both of which are entirely untrue - all I'm trying to represent is that I've had very good (and fast) results in experiments with a useful GUI interface, that's all.....
Last edited by RobustReviews; 09-24-2021 at 07:11 AM.
|
The following users thank RobustReviews for this useful post:
lollo2 (09-24-2021),
ThumperStrauss (02-01-2022)
|
09-24-2021, 07:22 AM
|
|
Free Member
|
|
Join Date: Mar 2013
Location: Italy
Posts: 676
Thanked 193 Times in 165 Posts
|
|
Quote:
however when you're working on about $6US/hour quite often for footage they're not practial
|
That's probably the key point!
I am putting all the needed effort on my videos, cause I do not have constraints.
If you need quick and acceptable results Topaz VEAI and Neat Video (with the right settings!) are a good option.
Edit:
Quote:
You've also used 'Arthemis' in the comparison of yours that I opened, it's very poor with VHS footage
|
None of Topaz VE AI models are adequate for VHS footage
|
The following users thank lollo2 for this useful post:
RobustReviews (09-24-2021)
|
09-24-2021, 07:29 AM
|
|
Invalid Email / Banned / Spammer
|
|
Join Date: Oct 2020
Location: London - UK
Posts: 568
Thanked 88 Times in 76 Posts
|
|
Quote:
Originally Posted by lollo2
That's probably the key point!
I am putting all the needed effort on my videos, cause I do not have constraints.
If you need quick and acceptable results Topaz VEAI and Neat Video (with the right settings!) are a good option.
|
That's what I'm trying to arrive at, before it's blown out of all proportion. We can run up to sixteen simultaneous transfers (across all formats from N1500 -> DigiBeta) and there are various levels of customer expectation/charge. We do this for business, not hobby as I gently try and remind people - that's quite a different notion when you can spend hours experimenting with your own footage. We work from budget transfers through to broadcast and everything in between, all have their own budget and expectation.
We're always experimenting with different options, but what we've dialled in certainly looks good to our eyes, but as I said, we're just experimenting at the moment.
Anyway, let's steer this back. I've found Topaz interesting and with some fiddling, it's possible to get some very good, fast and easy to operate results. AVISynth (et. al.) I know have sublimely good video restoration tools (and they're by the far the bulk of what we use) but let's not overlook some other stuff. I can also agree entirely with others that the deinterlacer is not beaten by QTGMC.
|
09-24-2021, 07:32 AM
|
|
Site Staff | Video
|
|
Join Date: Dec 2002
Posts: 14,176
Thanked 2,576 Times in 2,188 Posts
|
|
Quote:
Originally Posted by ehbowen
I get it that SD video is captured at 480i. But I've got a project in mind where I'd like to make some downloadable YouTube videos using some footage which was taped aboard a steaming battleship in the 1980s. Obviously it's not possible to do a re-shoot! I'd like to edit in some still photos at points in the video when I want to show something in higher resolution. I'd like to make the final video 720p, or possibly 1080p if someone wants a Blu-ray copy. So, if I'm going to be resizing and enlarging SD video from a VHS source (lossless captures with an AIW card), what's the way to make it look as good as possible?
|
Obviously you capture the tape at the max SD specs: 720x480/576, interlaced.
Then you process: deinterlace, upscale.
You must make wise choices here. You can, for example, deinterlace wrong, and actually lower the resolve/detail of the source. That's what happens when folks blindly use the QTGMC preset "slower". Slower blurs. The speed has nothing to do with quality, but rather the time needed to run the filters in the QTGMC preset. The fringes are generally bad, and slower is a fringe setting. You can opt for a better detail-retaining preset like "faster", or simply manually set QTGMC and not use presets.
Next comes the choice in scaling. Don't get bamboozled by marketing BS from the likes of Topaz. The scaler should only scale. Sharpening is a separate concept, don't cram them together like the Topaz goobers.
You must deinterlace first.
Then the generally sharpen and NR, scale (enlarge), and usually have to sharpen and NR again. When you scale, you emphasize noise and problems.
Quote:
Originally Posted by latreche34
In today's standards 720p doesn't make any sense anymore,
|
That's really not true. When you go from 480i/p content, it's best to scale to 720p. Then you can move up to 1080p if you feel the need. Trying to go 480>1080 in one go is often inferior, either slightly or largely. Many people do it, but it's rarely ideal.
Quote:
There is no native 720p displays left
|
That doesn't matter whatsoever.
Quote:
Therefore when you resize to 720p you get the first resize in software from 480 to 720 and the second resize during viewing from 720 to 1080/4K inside the TV or media player, since each resize is lossy you get double loss there,
|
Here's where you're making the mistake. Image theory. You're jumping too high, and not scaling in steps. The act of stepping doesn't inherently incur loss. In fact, it's often the complete opposite.
Quote:
on top of that Youtube algorithm classifies 720p as SD and applies harsh compression and limits the frame rate to 30fps.
|
To use a colloquialism that I'm sure ehbowen knows well: I don't give a rat's ass what Youtube wants. For the moment, I'm concentrating on the quality aspect. Youtube's shenanigans comes later. Youtube is nothing more than video blender anyway, in terms of "quality" (term used very loosely with them).
Quote:
What you want if you must resize is, first de-interlace with QTGMC, crop the edges down to the active video area while respecting the ratio 704:480 for NTSC, then resize to 1440x1080 square pixel, and upload to youtube, That's exactly what I do for youtube.
|
For Youtube use only, I do not disagree at all. Your video will be further raped by their destructive encoding, and quality is not the 1st priority (or even 2nd, 3rd ... way on down the list).
But he mentioned Blu-ray, which assume dual usage. So a better mastery copy is wanted, not merely Youtube. Make the Youtube version, to whatever ridiculous specs they insist on (for now), after the mastery copy is made for multi-use. Like the Blu-ray.
Quote:
Originally Posted by RobustReviews
This won't be popular, but we're starting to dial in some good results with Topaz Video Enchance.
|
Topaz software has tons of artifacts and weirdness going on. It's generally only impressive if you've not seen better Avisynth work. Topaz way oversharpens, and most people confuse sharpening with upscale. I could probably sharpen a 480p video as 480p, display it full size next to a truer non-oversharpened 1080 upscale, and some people would prefer the oversharpened mess.
Quote:
Those are other tools where you're mostly paying for the GUI, not the underlying filtering quality.
Quote:
if you're cautious, feed it good footage, and experiment with the settings.
|
That's really it. Those tools really are not made for SD to HD, but rather HD to HD. Those get very confused with videotape sources.
Quote:
Originally Posted by lollo2
For analog captures, there is nothing that Neat Video or Topaz Video Enhance AI can do that couldn't be achieved or beaten with Vapoursynth or Avisynth.
|
Correct. And there's many things that Topaz/NeatVideo/etc cannot do whatsoever, compared to Avisynth filters. The only thing going for those paywares is a GUI that makes rough NR/upscale easy for the novice. It's when novices claim those tools are "best ever" than they get a good thumping on the nose. Don't give me ground chuck, and insist it's steak!
Quote:
Once the trolls leave those threads, we actually have good discussion there. For example, poison's stacking script in the recent (and ridiculous) NV thread.
|
The following users thank lordsmurf for this useful post:
bar72 (08-27-2024),
dzikirobson (03-22-2022)
|
09-24-2021, 07:37 AM
|
|
Free Member
|
|
Join Date: Mar 2013
Location: Italy
Posts: 676
Thanked 193 Times in 165 Posts
|
|
Quote:
I can also agree entirely with others that the deinterlacer is not beaten by QTGMC
|
Topaz VEAI really needs progressive material. Do not even think (so far) to provide to it any interlaced stuff, results wll be disappointing
Quote:
For example, poison's stacking script in the recent (and ridiculous) NV thread.
|
poisondeathray (one of the AviSynth master from whom I learned a lot) simply recycled one of his scripts conceived to show that "Super resolution" was a BS to show to the troll the power of AviSynth versus NeatVideo.
As you said, filtering the trolls, those threads are a high value examples!
|
09-24-2021, 07:49 AM
|
|
Invalid Email / Banned / Spammer
|
|
Join Date: Oct 2020
Location: London - UK
Posts: 568
Thanked 88 Times in 76 Posts
|
|
Quote:
Originally Posted by lollo2
Topaz VEAI really needs progressive material. Do not even think (so far) to provide to it any interlaced stuff, results wll be disappointing
|
Agree entirely, the built-in DV/Interlaced presets are horrid, I'm not quite sure what they were developed for but nothing created by a human, that's for sure.
I use Proteus-6 (the one with the 'levers') which does allow some element of control, if you dial it in just right it can look very good.
But this wasn't a thread about Topaz, I was just reporting some results that I've had. 'Your mileage may vary' as I believe our North American cousins say.
|
The following users thank RobustReviews for this useful post:
lollo2 (09-24-2021)
|
09-24-2021, 08:22 AM
|
|
Free Member
|
|
Join Date: Feb 2018
Location: Houston, Texas
Posts: 339
Thanked 64 Times in 57 Posts
|
|
First, let me thank all participants for the good discussion last night while I slept. Some further amplification on my intentions in this project: First, the source material was taped aboard USS Missouri (BB-63) underway in the Pacific between Pearl Harbor and Long Beach. I was serving aboard the ship at that time, and the Navy has a "Tiger Cruise" program in which they sometimes permit guests to come aboard for the last leg of a long deployment. At the time we were returning from a six-month deployment to the Persian Gulf for Operation Earnest will in 1987, and so when we stopped in Pearl all of the sailors who could be spared took early leave and flew home to the mainland, while our guests flew out to Hawaii and took their berths for the trip back home to Long Beach.
My father flew out to join me, and he brought his VHS camcorder. After a little give-and-take he was allowed to use it onboard, with very few restrictions. He taped about five hours of material, and after a little digging in the closet I found it. I had found one of the tapes about two years ago (hence the 2019 capture date on the clips following), but I just dug the other two out very recently. Battleship New Jersey museum has a very popular string of YouTube videos produced around the ship, and I was struck by the notion that a similar series based on this material which was taped with the ship actually in operation might also find an audience. While YouTube is the primary distribution channel, I also wanted to be able to offer hard copies on Blu-ray at essentially cost of distribution and postage (I don't want to get into the legalities of having to research and obtain releases from other sailors in the videos, especially those who may be dead!) for those who want optimum quality.
I have an interview to go to across town and I need to get dressed and leave right away, but I wanted to upload a couple of brief samples. One is about a 45 second clip discussing the operation of the after steering gear in mp4 format (converted by Vdub2 FFmpeg), while the other is 400 lossless frames from the same clip as a direct stream copy from the original capture. Let's see what we can do with these. I do have Neat Video v5 for both VirtualDub and Resolve in my toolkit, but not Topaz.
|
09-24-2021, 09:14 AM
|
|
Free Member
|
|
Join Date: Dec 2017
Location: Norway
Posts: 1,685
Thanked 460 Times in 394 Posts
|
|
Quote:
Originally Posted by lollo2
Edit:
None of Topaz VE AI models are adequate for VHS footage
|
Not gonna comment on the quality of Topaz as I haven't used it, though if one is dealing with AI this point is very important. It needs the right training material to do a good job. An AI model trained on/designed for upscaling clean digital video won't be as ideal on noisy VHS. I'm sure in the future AI model could very well exceed anything possible with current avisynth filters on VHS footage if it had been trained on similar material with enough processing power thrown at it. It should be able to recognize signal patterns and such. Someone on the DD86 discord (zcooger) was playing around with some AI stuff a while back, this is just a single frame thing (for video one needs to consider multiple frames to avoid motion weirdness) but it's interesting nonetheless:
VHS_AI.jpg
Also should note, AI in the form of a neural network is used in the scaling part of the QTGMC deinterlacer, nnedi3. It's somewhat primitive in comparison to the topaz ai as I think it's only using it to detect/connect lines though it works reasonably well without causing weird artifacts.
|
09-24-2021, 10:19 AM
|
|
Free Member
|
|
Join Date: Dec 2015
Location: USA
Posts: 3,530
Thanked 601 Times in 552 Posts
|
|
Quote:
Originally Posted by lordsmurf
That's really not true. When you go from 480i/p content, it's best to scale to 720p. Then you can move up to 1080p if you feel the need. Trying to go 480>1080 in one go is often inferior, either slightly or largely. Many people do it, but it's rarely ideal.
|
Although this doesn't make any sense, I've further experimented with it and found that 720 is very destructive for the reasons I've explained.
Quote:
Originally Posted by lordsmurf
That doesn't matter whatsoever.
|
The display does matter a lot since you are at the mercy of its internal upscaler quality.
Quote:
Originally Posted by lordsmurf
Here's where you're making the mistake. Image theory. You're jumping too high, and not scaling in steps. The act of stepping doesn't inherently incur loss. In fact, it's often the complete opposite.
|
Resizing always incur a loss, It is mathematical approxiamtion and what's lost in the way will never be recovered, that's why it is prefered to do it once.
Quote:
Originally Posted by lordsmurf
To use a colloquialism that I'm sure ehbowen knows well: I don't give a rat's ass what Youtube wants. For the moment, I'm concentrating on the quality aspect. Youtube's shenanigans comes later. Youtube is nothing more than video blender anyway, in terms of "quality" (term used very loosely with them).
|
Me neither but this thread is about uploading to YouTube unless I missunderstood the OP.
|
09-24-2021, 11:02 AM
|
|
Free Member
|
|
Join Date: Mar 2013
Location: Italy
Posts: 676
Thanked 193 Times in 165 Posts
|
|
Quote:
Not gonna comment on the quality of Topaz as I haven't used it, though if one is dealing with AI this point is very important. It needs the right training material to do a good job.
|
My PhD was about neural network implementing back-propagation algorithm, so I fully agree with you.
Training/modelling it's not easy at all, but so far the achieved results on live video do not entirely satisfy me.
However I see a lot of development activity on the VapourSynth filters (VSGAN and similar), so I will keep an eye on it!
I use (excellent) Nnedi3 every (rare) time I wish to upscale and QTGMC every time I want to deinterlace, but it is a sort of NN, I would not call it AI.
Edit: the experiement you show is interesting, but that plastic look is (one of) what we dislike in Topaz VEAI. Not easy to find the good compromise...
|
09-24-2021, 11:03 AM
|
|
Site Staff | Video
|
|
Join Date: Dec 2002
Posts: 14,176
Thanked 2,576 Times in 2,188 Posts
|
|
Quote:
Originally Posted by latreche34
I've further experimented with it and found that 720 is very destructive for the reasons I've explained.
|
You can't approach 480>720>1080 the same way you would 480>1080. That probably would be destructive.
Let's look at the math:
720x480 > 704x480 ~ 640x480 1:1 4x3 AR
1280x720 ~ 960x720 1:1 4x3 AR
1920x1080 ~ 1440x1080 1:1 4x3 AR
A general rule for upscale, at least in the old days (and I see no reason why it's not still good advice), is to never 2x+ it. If it goes over 2x, it needs to be stepped, unless it's a perfect multiple. This pre-dates my digital video days, way back to basic digital photography.
That's probably why the Avisynth "rpow2" operates much in the same fashion, a 2x multiplier.
480 > 720 is a perfect 1.5
480 > 1080 is a less ideal 2.25
720 > 1080 is another perfect 1.5
That matters substantially with aliasing.
Quote:
Resizing always incur a loss, It is mathematical approxiamtion and what's lost in the way will never be recovered, that's why it is prefered to do it once.
|
Again, stepping doesn't necessarily incur loss. That's the same sort of philosophy that gets us H.264 deinterlaced captures in OBS (yuck), instead of lossless interlaced captures stepped to post-capture encodes when processed between steps. (Lossless has no loss, you say? Fine. Sub lossless for MPEG-2 in the example. Same thing. MPEG has loss, but direct to H.264 has more loss. MPEG>H264 best.)
Quote:
Me neither but this thread is about uploading to YouTube unless I missunderstood the OP.
|
It's about distribution, where Youtube is just one likely/potential avenue.
|
09-24-2021, 11:17 AM
|
|
Free Member
|
|
Join Date: Mar 2013
Location: Italy
Posts: 676
Thanked 193 Times in 165 Posts
|
|
Upscaling in steps is probably a better method because the math and because you can add a lighter sharpening at each step, producing less banding and artifacts then do it once on the final video, but for youtube upload is not worth imho
|
09-24-2021, 11:36 AM
|
|
Free Member
|
|
Join Date: Dec 2015
Location: USA
Posts: 3,530
Thanked 601 Times in 552 Posts
|
|
Rule of thumb in upscaling the higher the upscaling ratio the lower the resizing artifacts, It's math. From 480 to 1080 the ratio is 2.25 meaning that every 4 lines are rendered into 9 new lines, 480 to 720 has the ratio 1.5 meaning that every 2 lines are rendered into 3 new lines (yikes), So the resizing artifacts are far apart in the first case than in the second case not to mentien that the damage is repeated again when going from 720 to 1080 and just hope that the TV or media player has better upscaling than what Vdub can give with different resizing filters.
As you can see stepping always incur a loss because the approximation. Whenever an image is rescaled or resampled, it will suffer from one or more of the following artifacts:
Aliasing / moire
Loss of frequency response / blurriness
Ringing / halos
https://www.youtube.com/@Capturing-Memories/videos
|
09-24-2021, 11:45 AM
|
|
Site Staff | Video
|
|
Join Date: Dec 2002
Posts: 14,176
Thanked 2,576 Times in 2,188 Posts
|
|
Quote:
Originally Posted by latreche34
Rule of thumb in upscaling the higher the upscaling ratio the lower the resizing artifacts,
|
Huh?
With that logic, 4K would be better than 1080p.
That's ... no.
In a world without aliasing, I guess so. But with aliasing as a factor? Nope.
|
Similar Threads
|
Thread |
Thread Starter |
Forum |
Replies |
Last Post |
Device to capture 480i in 720p?
|
x77x |
Capture, Record, Transfer |
13 |
07-23-2017 01:19 PM |
Samsung BD 1080p, 1080i, 720p, 576/480p questions
|
Giana |
Videography: Cameras, TVs and Players |
1 |
01-02-2013 01:01 AM |
1080i or 720p for Youtube?
|
Winsordawson |
Edit Video, Audio |
15 |
10-28-2012 10:58 AM |
What else can I do to make my DVD's quality better ?
|
John |
Capture, Record, Transfer |
12 |
01-21-2011 02:15 AM |
Ripping one DVD to make another
|
Superstar |
Author, Make Menus, Slideshows, Burn |
11 |
08-09-2010 01:25 AM |
All times are GMT -5. The time now is 10:15 AM
|