Any chance of improving quality & size to a 352x288 video?

[tntroy]

I've mostly been watching and reading posts in this forum, since I joined a few weeks ago. Amazed at how much many people here know about video quality, size, converting, burning & all. I mean I've been playing around with videos in one way or another, with my minimal knowledge of the subject. For around 10 years or so. But compared to many of you all, and what you know? I may as well have started just yesterday, lol.

My question revolves around a bunch of videos I have, from back in the day. When 352x288 ( or 320x240 ) was a decent sized ( average ) video file. In both quality and specs. I guess since many people were still using dial up at that time. As well as slower versions of cable internet too. So a much larger, better quality file would take ages to download and upload? lol.

Is there any real way, with all the high tech programs out there? To both improve the size, say to anything larger then even 480x360? Without losing any quality? I mean I know the basics of video converting and have done the best with the program I've tried using ( programs that are at my knowledge level, lol. ) But even at only 480x360? You can already see some loss of quality. I guess because they also used very low settings back in those days as well? In other words specs ( properties ) such as 4:3, data rate 636KBPS & total bit rate 700KBPS & 30 frames per second. Not much you can really improve on, is there. Oh yes and they are WMV as well.

I've tried conversions using programs such as Format Factory, Windows movie maker ( which I liked since you can set many of your own specs on it. ) While Format Factory & the others come with preset specs. But lately I've been noticing if I make a video any larger then like 480x360 using Movie Maker? The actual video size doesn't get much larger? It used to, but not anymore. I've also tried a few of the online sites, to no avail for the same reasons. I've read about and downloaded programs like Avesinth, VirtualDub & some other one I cant remember the name of. But they were a bit above my knowledge level.

Probably a waste of my time and effort but any way. I like messing around with programs and learning how they work. And to see what I can do with them. I read a few other posts here about video with high glare from either the foreground or from the sides of the subject in the video. ( spots lights or sun glare ) Because I have a couple of short clips where bright sun glare is behind the subject as well. With the added problem that the subject is fairly far away from the camera. But I don't understand Chroma and a few of the other possibilities that could be used to improve that problem.

Again, probably something that not much improvement could be made of. Just figured I'd mention this and see if anyone wanted to give either of them a shot at improving! Thanks, if anyone reads this.

[Cornucopia]

Short answer, NO. Not really.

Longer answer, all forms of bitmapped resizing are guesses and thus lose quality. Downsizing (without zooming in) maintains the appearance of the same quality, just at a smaller size. But unless these are simple integer ratios (2:1, 3:1), there is still interpolation of much, if not all, of the image.
But upsizing is worse, as it doesn't so much do interpolation (double-ended) as extrapolation (single-ended), so there greater chance of error.
Into these equations comes the sizing methods (aka formulas), which vary from simple (point resize, nearest neighbor) to complex (lanczos, bspline36, etc incl. some custom, proprietary, or special purpose ones). If spec.purp. fits the criteria, some can be nearly lossless for that material, but that is not usually the case.
In general, simpler ones usually have more error or artifacts than complex ones, but there are exceptions and there are different kinds of artifacts which bother different people in different ways.

Then there is the scale factor. Larger ratios of scale usually imply larger ratios of guessing, made up data, and error (and thus loss). There is no free lunch.

Add to all that the likelyhood that you would be re-encoding to another lossy format and so you would incur additional loss ON TOP OF the loss due to resizing.

You could just view those images natively without zooming to full screen and it likely would appear as better quality, though smaller. It is actually the zooming to (usually a much larger) screen filling that in itself is a form of (up-?) resizing. Remember, back when those small sizes were popular, so too were smaller screen sizes for viewing them.

Scott

[tntroy]

Hey Scott Thanks for the reply, and with your more technical answer to my question? What it simply comes down to is this. When you used playing the same video, then pulling open say VLCs viewing screen larger then what said vid file opens to. You already see loss in quality then to. So all these converters are trying to do it "fill in the missing pieces" as you called it. ( Was info of course ) since everything we see digitally is just information. zeros and ones I suppose.

Conversion is, as you said. just trying to guess at what said info is now missing from original file. When making it larger ( or opening the viewing screen larger = same effect. ) The one unusual thing I have come across though which puzzles me. Especially now knowing that? I have dozens of vids shot & processed by the same photographer. Of the same exact size/quality video? yet some do convert better then others do. I guess the converter is just "guessing" better during those conversions? lol. At filling in missing info?

I guess it's like when you have 2 photos that when they are fully open. Are the same size from side to side, top to bottom? ( 1800x1200 ) Yet the KBs or MBs can be different. Depending on what info is in each photo. The dots or pixel count? Like if you have a subject in a certain outfit. In one photo, they are in front of say a plain back ground. While in the other photo, they are in a very detailed back ground. I forget now which photo would end up being large KB or MB wise but. You get my drift? Maybe the videos with a more plain background convert better? Or vice versa. Because there is less detailed info to fill in the missing spaces?

[KarMa]

One of the better upscalers out there while still being able to scale in realtime.

http://avisynth.nl/index.php/Nnedi3/nnedi3_rpow2

[Anonymous5394]

You cannot put detail which isn't there, but you can guess as a lot of these programs purport to do in enhancing upscaling using clever algorithm e.g Wondershare. You could try Wonderfox HD Converter Pro. There's a free version I think free HD video converter inputting a higher resolution e.g 360p and see how that works out.


https://www.videoconverterfactory.com/download.html

[johnmeyer]

You didn't say what frame rate your videos run at. Very often these old low-res videos run at only 10-15 fps, or less. You will want to fix that as well.

As already stated, you can't add detail, but with a good scaler you can smooth the jaggies that you'll otherwise get from the scaler in your playback software (all software scales video to fit whatever windows size you've chosen).

I did this same thing seven years ago. You can read about exactly what I did, including software and settings, and also including the final result. Here is the link:

Ideas for restoring early generation (really old) digital video

[Cornucopia]

@tntroy, much of the difference in quality is due to those choices in formula/algorithm.
Size is a different matter.
A photo that is 800x800 using 24bit per pixel in RGB will always be 15,360,000 bits of payload data (plus some small amount of overhead depending on the chosen formats/headers), no matter what is being shown - UNCOMPRESSED. The variation you are describing has to do with COMPRESSED encoding (lossy or lossless) and yes, that is very dependent on the content.
And then there is a difference between resolution and detail. Resolution has to do solely with the granularity of the sampling. Detail is related but also taps into our visual system and object recognition and so is dependent upon contrast, among other things. And, as has been overworked in another recent thread, detail isn't the same as sharpness.
But think of detail as "the unique info" in the file. The more the detail, the harder to compress.

Scott

[tntroy]

Wow, lots of good replies here today. Though I know from reading them all and then clicking on the one link to how someone improved quality a bit ( of these older vids ) My problem is exactly what I stated it is. Not having any where near the amount of knowledge of the technical aspects of video. The way you guys all do. Some of the terms being used are just like "OK". I have an idea what's being said at basic levels of what you are saying. But not really. I rate it on levels like this. I am at like 1st or 2nd grade level, while you guys are all at high school or even above. lol.

Most anything I've learned has come from experimentation. Try this, try that. Change these stats ( stats I understand any way. Because I know I even use the wrong terms, wording for specs now and then. ) I mean I keep using the word size when I mean resolution. Not the size of the video, as in MBs. I thought I mentioned the frame rate is 30 FPS. These vids were all shot & converted between 2004 to 2006. Only near the end, had they start trying using a 640x480 resolution. Right clicking on the files I guess only gives the basic info of the video.

The photographer was a novice at shooting video, never mind making a usable product out of the raw footage shot. I don't really think he actually knew a whole lot about that side of photography either so. I mean especially photographers & video people of today's scene know much more about either photo or video then back then. Digital wise, I should say. These guys all started using film & darkrooms so. Learning digital on the fly, so to speak. The "scene" I refer to is something that was known by Nonude models website girls/websites. The model in question is 18 & older in all the content. And doesn't ever get fully nude. So if anyone wants to take a stab at working on a clip from one of the vids. Or the entire video? I'll post a sample clip or full video. It's only recently I found out, even making clips of the original video. Kind of lowers the quality of the original video quality.

[jagabo]

Cartoons can be upscaled pretty well. They typically don't have small details, only sharp edges. It's possible to upscale those edges while retaining sharpness but not creating aliasing artifacts. See the video in this post (upscaled from DVD 720x480 to 1440x1080):

https://forum.videohelp.com/threads/387998-Ripping-a-DVD-in-order-to-deinterlace-decom...it#post2511590

Real world video doesn't upscale as well. If you try for sharpness like that things start looking very artificial. Everything looks plastic, people look like mannequins, etc.

I recommend you post a short sample of what you're looking to improve.

[tntroy]

Thanks for this suggestion. I'd like to try it, but a file scan shows it contains Trojan.Generic.bzzdf - I have no idea what kind of problems that might cause my computer. But since I had multiple problems in the past from installing "free" programs I never scanned? I never install anything that shows a positive ( or negative ) in a file scan.

You cannot put detail which isn't there, but you can guess as a lot of these programs purport to do in enhancing upscaling using clever algorithm e.g Wondershare. You could try Wonderfox HD Converter Pro. There's a free version I think free HD video converter inputting a higher resolution e.g 360p and see how that works out.


https://www.videoconverterfactory.com/download.html

[tntroy]

Thanks Jagabo. I was hesitant about posting any samples because of the content of said videos. Some people might find it offensive in some ways, since it is considered porn. Porn is Porn as in videos or photos of people having sex, in my book. Otherwise it is erotica in my book, it's in the eyes of the viewer to judge. As long as the subject in the video is an adult & consenting ( & making money from in this case ) I just figured Id take a stab as I've said before. To see if anyone else could make an better progress with them, then I had. Otherwise, it's time to trash them all.

I wasn't sure whether to use the deinterlace when converting or not. since I have no idea if it would actually help or further erode the quality of the original. I tried both and didn't really notice any difference. I guess simply because there isn't a whole lot can be done to improve a video this small in resolution. 480x360 looks ok. But again, you can already see some loss of quality, at that small of an upgrade. Going to 640x480 was obviously even more of a loss. I tried making my own resolution sizes on Movie Maker. But again, either something happened with that program installed on my computer. Or I don't know what ( I actually cant even use the Windows video player for some reason. Something screwed that program up on me. It doesn't work as well as it once did. Video quality seems to be darker then the original. And When I tried making a 720x480 with it? The resolution doesn't get much larger then a 480x360. Maybe it also negatively effected Movie Maker as well. Luckily I mainly use VCL or MPC to watch video

Cartoons can be upscaled pretty well. They typically don't have small details, only sharp edges. It's possible to upscale those edges while retaining sharpness but not creating aliasing artifacts. See the video in this post (upscaled from DVD 720x480 to 1440x1080):

https://forum.videohelp.com/threads/387998-Ripping-a-DVD-in-order-to-deinterlace-decom...it#post2511590

Real world video doesn't upscale as well. If you try for sharpness like that things start looking very artificial. Everything looks plastic, people look like mannequins, etc.

I recommend you post a short sample of what you're looking to improve.

[tntroy]

This is a clip of probably one of the least "revealing" videos by this "model".

[jagabo]

Here's a quick cleanup and upscale comparison. On the left is the original video upscaled with a simple bicubic filter -- typical of what most players or graphics cards would do. On the right is the video cleaned a little and upscaled better with nnedi3. I used QTGMC for its edge cleaning but it added some oversharpening halos. I also used its EZClean feature for some noise reduction (though I prefer other noise reduction filters). Both videos are cropped a little so they would fit in a 1920x1080 frame -- so most people will be able to see the whole picture pixel-for-pixel. I didn't bother including the audio. One could do a little better than this but this will give you an idea of what to expect.

[Bernix]

Hi,
some people mostly around anime, capture each frame to png and then let it resize with Waifu2x in batch mode. It is based on neural network, probably implemented in Avisynth don't know. Also not sure if there is any reasonable difference with Nedi 3 resizer.
It is for Anime, but not only. Four models I think. And it is very slow process. Waifu2x caffe (not only 2x resize as mentioned in name. Any size can be set) can also denoise if you wish. GPU CPU support (Cuda).
I think there is lot about it on doom9


Bernix

[tntroy]

WOW, jagabo! That's Amazing, especially at the size you enlarged it to. I could have seen it looking like that at like a 640x480 maybe. But never at a 1920x1080. If I could only perform even half that magic at the size. I'd be happy as a pig in mud. lol. was this using Avisyth? With all these filters and upscaling an all? If so, do they have a manual of any kind for it? A friend of mine will flip out, to see this result. He has DVD's of this model he wishes he knew how to "clean up" and improve quality of. Not that DVD is bad, but compared to today's high def videos? They are like comparing this original 352x288 to a good quality 720x480. ( The subject in this vid retired, so to speak in early 2009. ) My computer probably isn't powerful enough to handle the kinds of converting this improvement took though. Even if I knew how! For its time, it was one of the best but.

Here's a quick cleanup and upscale comparison. On the left is the original video upscaled with a simple bicubic filter -- typical of what most players or graphics cards would do. On the right is the video cleaned a little and upscaled better with nnedi3. I used QTGMC for its edge cleaning but it added some oversharpening halos. I also used its EZClean feature for some noise reduction (though I prefer other noise reduction filters). Both videos are cropped a little so they would fit in a 1920x1080 frame -- so most people will be able to see the whole picture pixel-for-pixel. I didn't bother including the audio. One could do a little better than this but this will give you an idea of what to expect.

[jagabo]

The AviSynth script I used (minus the side by side stacking) was fairly simple:

AviSource("Vid sample.avi", pixel_type="YV12") 
ColorMatrix(mode="rec.601->rec.709") # convert to HD colors
vinverse() # reduces some residual combing artifacts
QTGMC(InputType=1, EZDenoise=2.0, DenoiseMC=true, Sharpness=0.7) # cleans up a lot edges, minor noise reduction

# upscale about half way to 1080p
nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=720, fheight=540)
aWarpSharp(depth=3) # smooths edges and sharpens
Sharpen(0.3)

# upscale to 1080p
nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=1440, fheight=1080)
aWarpSharp(depth=5
Sharpen(0.5)

I originally used LWlibavVideoSource() to read the WMV file but that was producing problems with the out-of-order processing done in the script (QTGMC may not request frames linearly). So I first converted the WMV file to an AVI file using the lossless UT VIdeo Codec. The script for that first step was:

LWlibavVideoSource("vid sample.wmv")

I opened it in VirtualDub, selected the compression codec, and saved as AVI. Then proceeded with the first script. I produced the MKV video with the x264 command line encoder (it can work with AviSynth scripts as input). You can use any editor/encoder that accepts AviSynth scripts as input. AviSynth, MeGUI, etc.

Starting with a DVD source will require slightly different filtering but will probably give better results. Your computer is a little slow but should have no problems with these scripts once you have AviSynth set up.

AviSynth is a little difficult to get started with since it's controlled via plain text scripts. "Plugin" filters are written by different people all over the world. And many of the filters require someone else's filter(s) to work. Sometimes one person's FilterB() may require another persons FilterA() version 1.0. But the person who wrote filterA() is now on version 3.0. So you have to find version 1.0 somewhere. QTGMC is one of the worst in this respect.

[tntroy]

Dumb questions from a newbie. I've tried to use AviSynth & Virtual Dub before, but had the same problem I read others here also had with virtual dub. How can I convert the wmv to an AVI with it. If I cant get Virtual dub to open the WMV? lol. I tried converting it with the Format Factory I have, first as an AVC 264, then as an X-vid. But neither worked. I even installed the FFDshow 64 pack & still nothing. I installed the FFDshow 64 pack because I read in another thread here, where another newbie asked how can he get an AVI file, Virtual Dub will open?

[jagabo]

Use VirtualDub2 instead of VirtualDub.