Giveaway of the Day Forums

1. 

   mikiem2
   Moderator

   extremetech[.]com/extreme/338403-what-video-ai-upscalers-can-and-cant-do

   1st off this is a Long read, has some info that should be useful if you're thinking about trying AI video enlarging, But, treat it as one individual reporting their experiences, not as some sort of How To or guide. I don't mean to shortchange the amount of work the author's done, but from his workflow it seems his background in image & video editing and manipulation might be somewhat limited...

   AI video enlarging has different models available that determine what the results look like, and one of its more tedious aspects is trying to find the best model for the video you're working with. And that best model may vary from one scene to the next. The obvious solution is to split the video into separate clips, enlarging scenes separately, then putting the clips back together in your video editor. While the problem of different models for different scenes is addressed, processing clips individually doesn't seem to be.

   There's also not much of any discussion of video formats. The author talks about his saving enlarged video as individual images, and while that can work, IMHO it's definitely Not the way to do it. I think it's Much more efficient to use a lossless codec like Ut Video -- there are several -- though in fairness he does mention Apple's ProRes. Another aspect of codecs should also be mentioned -- working with video, software must decode the source, and for that the source format really does matter. Unless the source is using an editing format, like the mentioned UT video, some software may add &/or drop frames [both are possible in the same video (!)] so audio sync is almost impossible. You need to test your complete workflow, looking for this sort of problem before you start a project, so you can for example convert the video to an editing friendly format first if you need to. This part of things can get pretty involved, so be prepared to research.

   One other omission that I think should be addressed concerns the author's use & mention of AviSynth & VapourSynth... They're both *frame servers*. If an app cannot open a certain video, you can use a compatible frame server instead -- the app requests the video from the frame server, which feeds it frame by frame to the app. AviSynth & VapourSynth do more than just feed the video however -- both can be scripted to perform all sorts of manipulations. There's a good comparison of the 2 here: video.stackexchange[.]com/questions/28548/avisynth-vs-avisynth-vs-vapoursynth-which-one-should-i-choose

   You'll have to do some research too on these frame servers, and optionally which GUI to use with these command line apps. Google on AviSynth, AviSynth+, &/or VapourSynth and the video editor you want to use for directions on how to get video output into your editing app. If this all seems rather daunting, search the software at videohelp[.]com for VirtualDub... Like AviSynth, VirtualDub was born a Loooong time ago, when people like me were recording broadcast TV and saving it to Real & Windows Media formats, and sometimes VCDs. Like AviSynth, coders have made efforts to update the old software, with a version of VirtualDub as recent as 2020. Because VirtualDub & AviSynth were both designed for generally the same tasks, working with & cleaning up broadcast video recordings, their filters had lots of overlap, and you might well get the results you need from VirtualDub without the hassles of learning AviSynth or Vapoursynth.

   Posted 2 years ago #
2. 

   mikiem2
   Moderator

   Enlarging DVD video

   This involves some special stuff that won't concern anyone enlarging 480p, 720p, or 1080p video, so I wanted to talk about it separately. I'll start by saying that if a movie or whatever is available on Blu-ray or streaming, you're better off biting the bullet & buying it in HD or 4k -- this is especially true if the DVD you want to enlarge was released after Blu-ray video became available, because the DVD's quality will be artificially reduced to discourage exactly what you want to do.

   That out of the way, before LCD displays became available, everything used CRTs [Cathode Ray Tubes]. These worked by shooting a beam at the phosphor-coated backside of the face of the display, making the phosphor glow. Today you talk about pixel density when comparing the finest resolution display panels are capable of -- back then it was all about how many scan lines the CRT had... the higher that number, the better the *potential* picture, depending on the number of lines in the video you wanted to watch [if the video had 300 lines & the TV had 600, the TV *might* upscale the picture, e.g., similar to many 4k TVs showing 1080p video]. Because the phosphor coating continued glowing for a short time, that electron beam would skip every other line, scanning top to bottom, then scan the ones it missed. That effectively *simulated* a doubled frame rate [fps = Frames Per Second], while not having to store & deliver that extra data. Each half of a frame is called the Field, and whether the video itself was intended to be displayed with odd or even scan lines taking priority matters. This is called Interlacing, and it's still done occasionally with 1080 video.

   That very last part is important. There is no good way to remove interlacing or to enlarge interlaced video, because the secondary field has a very incomplete picture that will not align with the primary field in the frame. Depending on the video, you *may* have better results enlarging it to 1080 interlaced output. The alternative is to try various de-interlacing filters &/or methods, and hope that you wind up with something usable.

   While TV cameras recorded interlaced video, the alternative was film, which does not have interlacing -- this is called Progressive video, and is the "p" in 1080p. Almost 100% of movies on DVD are progressive video, and quite a few TV shows too. Film is normally shot at 24 fps, so to be compatible with a TV that expected 29.97 fps interlaced video, certain frames are shown twice in something referred to as 3-2 pulldown. Sometimes they just faked it, sending 24 fps video instead of PAL video's 25 fps to PAL TVs. Removing those duplicates is called Inverse Telecine, and if/when the video uses pulldown, should be performed before doing anything with the video. Unfortunately there are some DVDs that used both video at 24 fps & 29.97 fps, and the field order may even change in different scenes. All you can do is try to output at 1080 interlaced or process the video scene by scene, doing what needs to be done as you encounter it.

   This next part is [sadly] harder to understand because it seems, well, so arbritary. Again going back in history, with analog TVs [basically everything before HD], it was all about waveforms and wave theory. I'll spare you the math behind it and just give you the [perhaps surprising] result: PAL & NTSC analog broadcast & DVD video Do Not have or use square pixels. Again skipping the math, a 720 x 480 video frame is meant to be displayed pretty close to 640 x 480 on your PC, laptop, or really anything with a display nowadays, like your phone. But, That was to fill the screen on an analog TV with its ~4:3 aspect ratio. 16:9 widescreen video on a DVD is still the standard 720 pixels wide, BUT, stretched by the DVD player before the signal's sent to the analog TV. How much is it stretched? There is no standard (!). So, if you want to display DVD widescreen video on a current display you can use a DVD player [hardware or software], which will stretch the video to approximately what it should be, but that won't won't work after you resize it -- if you want to enlarge that video you have to resize it to roughly 16:9 And then enlarge it.

   One trick that you can use to determine the approximate frame size of 16:9 DVD video is to take a screen shot while playing it using a software player on your PC/laptop -- opening that image in an editing app you can get the exact size of that frame. [The reason for saying *approximate* frame size is that as above there's no standard, so the frame's measurements can vary depending on what player software you use.] Most video convertors can of course make that size adjustment for you, but they usually cannot save the output using a lossless codec, which is what you want -- a video encoder that is not lossless will reduce the quality, which you definitely do not want prior to enlarging the video.

   Finally, if you want to enlarge DVD video you have to deal with the audio too. You'll usually be happier with the results if you convert the audio to .w64 [wave 64], apply level compression, because the audio was probably designed for a theater's sound system, and then check sync, often nudging it a bit towards the beginning or end. You might also have to split the audio track wherever/whenever you split the video for processing & adjust it there too.

   Posted 2 years ago #

RSS feed for this topic

Reply

You must log in to post.