Would this be resizing?

[hello_hello]

One "advantage" of stuff shot with 16mm film, is it generally has nowhere near 1080p worth of picture detail, so it can be resized down quite a fair bit. Today I was playing around with a 1080p sample while resizing down to 540p for encoding and then back to 1080p for playback. There's a few shots where a wire fence can be seen through a wire fence, so I thought I'd compare the resized version to see how well the fence resized. As it turned out, the resizing didn't mess with it enough for me to care, but I noticed the original 1080p version appears to have been resized at some stage, maybe badly. According to IMDB this was shot in 16mm, so I'm curious as to how the 1080p version ended up like this. The video is progressive. If you look closely at the wire fence in the foreground it appears to be suffering from some sort of resizing jaggies. I'm not sure if it'll show up if the forum software resizes the images here, but it will if you look at them full size in a separate tab.

A section of the original 1080p video:


After QTGMC in progressive mode, resizing down to 540p, a little LFSMod(), then back to 1080p on playback. Sometimes this type of effect when using QTGMC on progressive video would be undesirable, but in this case the side-effect was more beneficial.


Anyway..... what would cause the jaggies in the original 1080p video? Would it be anything other than bad resizing?

[TorBru]

The original frame looks interlaced - maybe you should try deinterlacing the original? To my eye the second (resized) frame looks far better

[-Habanero-]

If 35mm is 4000p, 16mm should be far above 1080p.

[jagabo]

If you zoom into the image you'll see that every other line is nearly an exact duplicate. So it was likely 540 lines tall and resized to 1080 lines with a point resize. The small differences are probably compression artifacts. Probably poor deinterlacing.

[hello_hello]

If 35mm is 4000p, 16mm should be far above 1080p.

This is the best link I can find at short notice, but the BBC considers 16mm to be standard definition. I don't know if they've relaxed the 16mm rule these days.
https://recombu.com/digital/article/bond-director-calls-on-bbc-to-relax-hd-16mm-film-ban_M11447.html

I'll confess I'm somewhat skeptical about 35mm being much higher than 4k. Or even the equivalent of 4k a lot of the time. I'm not a film maker and know very little about film, but if it's got such a high resolution I don't understand why so much of it ends up on Bluray without 1080p worth of picture detail. Maybe much of it is lost in the transfer process. I don't go to the Cinema much myself, but even back in the film projector days I don't recall being blown away by film's resolution. I've always thought it tends to look grainy.

The main problem with 16mm and resolution appears to be noise. I have seen 16mm that looks good, but mostly it's noise to picture detail ratio probably isn't good enough to be classified as high definition.
ie from season two of Burn Notice.
There's cleaner shots and there's dirtier shots, but it jumps out as being 16mm to me and IMDB seems to confirm that.





If you've watched the Walking Dead, from which the screenshot in the first post was taken, you'll know how noisy it is. It's shot using 16mm.

If you zoom into the image you'll see that every other line is nearly an exact duplicate. So it was likely 540 lines tall and resized to 1080 lines with a point resize. The small differences are probably compression artifacts. Probably poor deinterlacing.

You're probably right, and after looking closer at other sections of video, there's similar artifacts. Where there's credits (you can see it at the edge of the car bonnet and along the wipers):



And the harder I look the more I find (resized to double the original resolution with Spline36):



That particular video is a few years old now and my recollection is it came straight from itunes, so I wondered why it wasn't simply progressive and why it'd ever need to be de-interlaced.

As it turns out, it may just be something that was only inflicted on that particular episode, and it just happens to be the episode I was using for testing, so I noticed it....... where's the law of averages whenever I need it....... but a quick look at some other episodes seem to indicate they appear noisy and 16mm, but without the jaggies.

A small section of 1080p from another episode. Original resolution (ie not resized).

The original frame looks interlaced - maybe you should try deinterlacing the original? To my eye the second (resized) frame looks far better

I think jagabo is right and it's already been de-interlaced, but I'll try QTGMC in de-interlacing mode rather than progressive mode later on, although progressive mode seems to be working well.

[jagabo]

When you enlarge images to analyze detail you should use PointResize(). That makes it very obvious where pixels start and end.

PointResize(16x):


Spline36Resize(16x):

[-Habanero-]

I'm not a film maker and know very little about film, but if it's got such a high resolution I don't understand why so much of it ends up on Bluray without 1080p worth of picture detail.

Because they're mastered at the resized 1080p resolution with all those crappy lowpass filters they use to smear grain, noise and other artifacts.

Check for Blu-rays that are advertised as being "Mastered in 4K", they have full detail at 1080p. This is because they do all their filtering and shit on it at its full resolution so when it's resized to 1920x1080 it still looks fully crisp.

A Blu-ray mastered at 1080p will have 720p detail, something mastered at 720p will have crisp SD detail and so on.

I don't know why that article says 16mm = SD but it is not true. 16mm is still far better than 1080p. But there are more aspects to film than its size, there is also quality of the material. Lower quality stock will mean less detail and faster decay which brings more noise and other problems. There's also the fact that film dots are not perfectly squared and aligned like digital pixels. So there's a bit of a "variable resolution" within the resolution.

[hello_hello]

I'm not saying you're wrong.... I certainly haven't seen every video/film out there.... but I'm yet to see a Bluray taken from 16mm that has a lot of picture detail. At least not 1080p's worth.

Have you actually compared "Mastered in 4K" Bluray video to standard Bluray video yourself? Does it really contain more picture detail? I'm not sure I've ever paid attention to the "Mastered in 4K" thing so I don't know if they do or don't contain more detail. I'd be interested to look at a sample that can't be downscaled at all without losing detail if you happen to have one.

[jagabo]

That particular video is a few years old now and my recollection is it came straight from itunes, so I wondered why it wasn't simply progressive and why it'd ever need to be de-interlaced.

I bet it came from an interlaced SD PAL source and was crudely deinterlaced and upscaled. I seem to recall somebody else posting a sample from itunes that was deinterlaced the same way.

[Brad]

If 35mm is 4000p, 16mm should be far above 1080p.

I'm assuming you're equating 35mm with 4K? 4K is not 4000p, it's 4096 in the horizontal dimension with varying height according to the aspect ratio: https://en.wikipedia.org/wiki/4K_resolution#Resolutions

Have you actually compared "Mastered in 4K" Bluray video to standard Bluray video yourself? Does it really contain more picture detail?

When the old master was inferior, like Spider-Man. Some "Mastered in 4K" BDs are basically just a higher-bitrate encode with different colors... like Spider-Man 3. The original BD of that was already sourced from the 4K digital intermediate.

I'm not sure I've ever paid attention to the "Mastered in 4K" thing so I don't know if they do or don't contain more detail. I'd be interested to look at a sample that can't be downscaled at all without losing detail if you happen to have one.

Try those two, but you should struggle to downscale any good Blu-ray. Like Ben-Hur.

[-Habanero-]

I'm assuming you're equating 35mm with 4K? 4K is not 4000p, it's 4096 in the horizontal dimension with varying height according to the aspect ratio: https://en.wikipedia.org/wiki/4K_resolution#Resolutions

You assume wrong, the resolution full UHD has been set to is 7680x4320 which they claim is the full and final resolution of 35mm film. Whatever transitional shit they have set is exactly that, transitional. It rakes in money, keeps morons spending and by the time they reach the horizon they'll either abandon 35mm film or ascend out of the box to 3DTV.

[hello_hello]

So nobody's got any samples?

[-Habanero-]

So nobody's got any samples?

Not at the moment, sorry. Anyone have 4K remastered Blu-rays on hand?

[Brad]

So nobody's got any samples?

I thought the screenshots I linked to would suffice for your downscale test.

[hello_hello]

I couldn't work out how to download the 4k version directly, so I ran it full screen and took a screen shot. To be honest I knew before I started it'd downscale without losing detail.

Mastered in 4k at 1080p:



Resized to 720p and back in Irfanview, which means it used lanczos resize, then I used Irfanview's standard sharpening filter to apply a very small amount of sharpening to compensate for the not so sharp resizing. Normally, I'd do it a little differently when re-encoding video, but it sufficed for a quick test. The overlaid text in the top right hand corner took an obvious quality hit, but the picture itself..... not so much.
As sometimes happens when downscaling, if a sharp resizer is used along with a sharp upscaler, the 720p version can end up looking a little sharper than the source. Not that there's much difference here, but the downscaled/upscaled version is just a tad sharper.



If the Ben-Hur screenshot you linked to is any indication, it's so devoid of fine picture detail I think I could downscale the video to 540p and back and running fullscreen you'd probably not be able to pick it from the 1080p version.

[hello_hello]

Here you go. Ben-Hur, downscaled to 540p and back again, same method as before.

1080p:

[Attachment 33062 - Click to enlarge]

1080p to 540p to 1080p:

[Attachment 33063 - Click to enlarge]

[Brad]

I don't know what kind of screen you're using to view these or from what distance, but there is a massive difference between those two Ben-Hur images, and the downscaled Spider-Man screenshot also shows significant losses. If I didn't know better I would think you were trolling me here.

[hello_hello]

I don't know what kind of screen you're using to view these or from what distance, but there is a massive difference between those two Ben-Hur images, and the downscaled Spider-Man screenshot also shows significant losses. If I didn't know better I would think you were trolling me here.

If I didn't know better, given I started this thread, I'd assume any reference to me trolling you indicates you're an idiot with a vivid imagination.

Can you offer something other than a meaningless "massive difference" generalization? Something a little more specific?

I'm viewing these on a 51" Plasma sitting next to my desk, and opening the images in Irfanview and switching between them, the Ben-Hur downscaled version appears to be a tiny bit blurred compared to the original, which is hardly surprising given I reduced it to 540p, while the Spiderman downscaled version looks a tad sharper over-all.

As I previously explained, I simply used the image viewing program I have installed for the resizing, and normally I'd use better resizing in Avisynth, but even though it's probably sub-optimal I think you still need to check your monitor's settings and disable it's placebo option, or at least dial it back quite a bit. A "massive difference". I don't think so.

Check the file sizes for the original 1080p screenshot and the downscaled/upscaled version. For the Spiderman screenshots it's 2.91MB vs 2.85MB. If that's not a fair indication as to how little difference there is between the images in respect to how much information there is to compress, I'm not sure what is. Or maybe those "significant losses" somehow only equate to a 0.06MB reduction in file size. Is that only something like 1.5% worth of significance, or is my math bad?

[Brad]

If I didn't know better, given I started this thread, I'd assume any reference to me trolling you indicates you're an idiot with a vivid imagination.

What's up with the personal attack? I'm sorry if I offended you with that comment. I found your previous post genuinely surprising.

Something a little more specific?

Yes, but does it matter? Your mind is made up.

Some specifics: the rings on the central seated character's hands have tiny details that are gone in the 540p shot, the man in grey's right arm has two white lines that are aliased, the fine vertical striping along his clothes is blurred away, his face no longer stands in sharp relief against the out-of-focus men and scenery behind him.

I think you still need to check your monitor's settings and disable it's placebo option, or at least dial it back quite a bit. A "massive difference". I don't think so.

Not sure what to tell you. I'm limited to a laptop screen at the moment, but I need to be over 3 feet away from it before the 540p shot can begin to pass for the original.

Check the file sizes for the original 1080p screenshot and the downscaled/upscaled version. For the Spiderman screenshots it's 2.91MB vs 2.85MB. If that's not a fair indication as to how little difference there is between the images in respect to how much information there is to compress, I'm not sure what is.

I believe sharpening does make things less compressible. I'm not particularly interested in arguing about the degree to which lossless compressibility relates to how our visual system sees detail, though.

[hello_hello]

Sure, I've made up my mind. I used my eyes to determine how much detail was lost.

Yes, the Ben-Hur screenshot blurred a little due to the fact I resized it to 540p and back. No doubt about that. I made up my mind and mentioned it a few posts back. What you're describing is probably more the effect of not so optimal resizing. A bit of contrast was lost along with a bit of sharpness, but I'm still certain the Ben-Hur screenshot doesn't have 1080p worth of picture detail. I'd very rarely downscale that much anyway because surprise, surprise, it's too low a resolution. It was just supposed to be a quick way to show 1080p video doesn't always have 1080p worth of picture detail, not a demonstration of downscaling perfection.

I'm not offering this as an example of detailed magnificence, it's just the first 1080p video I opened on my hard drive and a fairly random frame from it, but comparing it to the Ben-Hur screenshot it's night and day, picture detail wise. The Ben-Hur screenshot is almost cartoon-like, by comparison.



I had a closer look at the Spiderman screenshots. It was only downscaled to 720p which is more realistic. Once again, I'm just resizing with Irfanview's built-in resizing, but you can see the 720p downscaled version did lose a small amount of fine detail, once you upscale it enough for even the 1080p version to look fairly horrible. Once again that's probably mostly due to the resizing, but if there's significant detail losses at normal resolution, I still can't see them. It wasn't supposed to be a demonstration of downscaling perfection, just a rough experiment to see if it's fill to the brim with 1080p worth of goodness. I still don't think it is.

1080p:


720p:

[hello_hello]

Where'd everyone go? I was hoping for some 16mm screenshots chock full of picture detail at 1080p. Apparently not.....

Here's a couple of comparison screenshots I made today.

Burn Notice, Season two, 720p, shot with 16mm film.


Burn Notice, Season five, 720p, shot using a Red One MX 4K digital camera.


That's only a 720p comparison.
Is all the 16mm goodness really lost in the transfer process? Well, except for the noise. That seems to transfer well.

[Brad]

[Back from vacation/etc.]

To me it looks like his face is actually out of focus in that Season 2 frame.

I'm not offering this as an example of detailed magnificence, it's just the first 1080p video I opened on my hard drive and a fairly random frame from it, but comparing it to the Ben-Hur screenshot it's night and day, picture detail wise.

I guess you're only interested in facial detail? His necklace is also sharply detailed, but most of his clothing (like the zipper) is soft. The Ben-Hur shot has tons of detail throughout the frame, but not much on most of the faces.

[hello_hello]

The main reason I tend to focus on facial shots when checking for detail is because to me it seems like a good way to determine how much there is. If you can see detail in someone's skin, such as lines or blemishes etc, then the image probably has more fine detail than if you can't. I'm up for any sort of comparison but I'm not sure shots of more distant objects or outdoor scenes etc tend to be as good an indicator. Sharper images can sometimes appear to have more detail just by being sharper, and focus obviously makes a difference, so I tend to look for detail in close up shots, as in-focus as I can find, of objects that should have a lot of detail. Peoples faces seem to fit the bill.

When you say the Ben-Hur screenshot has tons of detail around the frame, I can only assume we're looking at completely different Ben-Hur images.
I've got the 1080p version and the downscaled 540p version both displaying fullscreen on my TV at the moment, while I switch between them, and I'm not seeing tons of detail disappearing when I switch to the 540p version. Some sharpness yes, the guy in the chair, his clothing and the same for the guy standing behind him, they're a bit blurred in the 540p version compared to the original, but the rest of the frame.......

Maybe you could explain what I'm not seeing. Here's a screenshot from a video I'm watching on my TV as I type. Not a close-up this time. It's 720p, upscaled to 1080p by MPC-HC. How does the mastered in 4k Ben-Hur image at 1080p have more detail? I just don't see it.

[poisondeathray]

I'm with vaporeon800 on the Ben-Hur example - there is a big difference there (maybe not "massive", but big)

[hello_hello]

The Walking Dead, 1080p, 16mm:



Fear The Walking Dead, 1080p. I'll go out on a limb as assume they're using digital cameras for this one. I haven't checked.



Sure, there's less detailed and more detailed shots in each format, but you only have to watch a few minutes of each to know the 16mm format is being left for dead in respect to sharpness and detail.... and lack of noise.

[hello_hello]

I'm with vaporeon800 on the Ben-Hur example - there is a big difference there

I still maintain most of the difference is sharpness related due to the way I resized, and that reduced the detail as a result. You'd kind of hope there'd be a difference though, given it was 1080p, mastered in 4k, and I downscaled to 540p and back. Maybe I'm missing the obvious.... I can see the guy in the chair and his clothing and jewellery were blurred, and I can see the guys behind him took a blurring hit, and the clothing the guy on his left is wearing lost some detail as a result, but once you look further left or right of those three central figures.... any detail differences seem to be fairly small to me, considering it was 1080p, mastered in 4k, and I resized it down to 1080p and back.

[Brad]

I'm not sure what you expect I can say about the one with Voight given that the content is so different.

How about if we back up and compare some CGI?



Do you agree that the 720p frame is markedly worse in the areas of the frame where there is fine detail to begin with? The treetops that are only a couple pixels wide get blended into one another, as do the delineations between the mountain's rocky surface and the snowcap. The tiny character's gaping lower jaw is faintly visible in the original, but I can't make it out in the downsize.

If you see the degradation there, then some live action scenery:

but once you look further left or right of those three central figures.... any detail differences seem to be fairly small to me

There's limited depth of field here: only objects that are within a certain distance from the camera are in focus, so those are the only places you should be looking for fine detail. The guy with the... wrist things... to the left is also in focus and his clothing is detailed, but parts of him are moving.

Sure, there's less detailed and more detailed shots in each format, but you only have to watch a few minutes of each to know the 16mm format is being left for dead in respect to sharpness and detail.

It's not a fair comparison for either format. Both of those frames are horribly compressed (iTunes?). The BD for The Walking Dead isn't just a smear of vaseline. I would agree that even the BDs don't have "1080p detail" from what I've seen, though.

[poisondeathray]

Everyone has different perceptions - some people think youtube looks great at 1080 - with super clear picture , lots of detail... I'm just calling it like I see it. It's not necessarily right or wrong

Now clearly some of the retail BD screenshots posted in other threads demonstrated very little difference when down and up scaled, but that Ben Hur example - there is a big difference. If you think those differences here are "small", I'm curious at what you think an example of "big difference" would be for real content (not test patterns)

Properly mastered 4K just means you have the ability theoretically get maximum resolution and detail at 1080p because of oversampling . But that doesn't mean you will on the final deliverable. Sometimes the deliberate choice is made for soft focus, sometimes production error limits effective resolution, sometimes heavy grain limits effective resolution

[jagabo]

Do you agree that the 720p frame is markedly worse in the areas of the frame where there is fine detail to begin with? The treetops that are only a couple pixels wide get blended into one another, as do the delineations between the mountain's rocky surface and the snowcap. The tiny character's gaping lower jaw is faintly visible in the original, but I can't make it out in the downsize.

Yes. And there are ~2 pixel wide oversharping halos around all the sharp edges from the sharp resizer. A more neutral resizer would have left the downsized/upsized picture even blurrier.

[hello_hello]

I'm not sure what you expect I can say about the one with Voight given that the content is so different.

How about if we back up and compare some CGI?

That's not backing up. That's turning left and travelling in a whole different direction.
The debate initially started when I said 16mm doesn't have 1080p worth of picture detail. I've not seen anything to prove me wrong there. I also questioned whether mastered in 4k Bluray always has 1080p worth of detail. I'm still not convinced it always does, although I'm now leaning in the "more likely to" direction. If you look at the screenshot comparison for your Lawrence Of Arabia pic, I'm not sure there's a great deal of difference between the 50th anniversary Bluray and the mastered in 4k version. The colours are different which makes it harder to tell, but to me the 4k version looks like it's been sharpened a bit. Maybe that's the result of the oversampling poisondeathray mentioned, or maybe it's just my imagination......
Anyway, having said all that, I couldn't downscale the mastered in 4k screenshot without losing detail. Mainly the detail in the road.

Do you agree that the 720p frame is markedly worse in the areas of the frame where there is fine detail to begin with? The treetops that are only a couple pixels wide get blended into one another, as do the delineations between the mountain's rocky surface and the snowcap. The tiny character's gaping lower jaw is faintly visible in the original, but I can't make it out in the downsize.

I'd certainly agree.... if your definition of "markedly worse" constitutes some blurring of very fine detail you can see in a still image on a 51" Plasma from a couple of feet away, once you make an effort to look for it. "Markedly worse" might be a bit of an exaggeration, but yes I can see the differences you've mentioned. I could also make a 720p version look mostly sharper than the 1080p version with a bit of playing around in Avisynth, despite the need to convert to YV12 to do it, except for the tree tops. Once they were blurred the sharpness/detail was lost.

Possibly one of the reasons I don't usually see a loss of fine detail when downscaling is because mostly I apply some sort of noise filtering during the process (I don't like noise), and noise filtering tends to remove some detail no matter how good it is. Or the noise gives the impression of detail that's not really there. I'd not worry too much about blurring a few microscopic treetops I wouldn't notice at normal viewing distance when the pictures are moving. If I don't use noise filtering, I'm less likely to downscale, and hopefully noise filtering will become a thing of the past as film fades away and everything's shot with digital cameras.

There's limited depth of field here: only objects that are within a certain distance from the camera are in focus, so those are the only places you should be looking for fine detail. The guy with the... wrist things... to the left is also in focus and his clothing is detailed, but parts of him are moving.

Yes, and unless you cherry pick tightly focused frames where there's no movement, including the camera, that's a big reason for being able to downscale without an apparent loss of detail much of the time.

It's not a fair comparison for either format. Both of those frames are horribly compressed (iTunes?). The BD for The Walking Dead isn't just a smear of vaseline. I would agree that even the BDs don't have "1080p detail" from what I've seen, though.

That's true, although I've had the same debate before. Yes, they both came from iTunes, and yes the 16mm shots probably needed a much higher bitrate (the digital shots don't look overly compressed to me), but does that mean 1080p pictures always have 1080p worth of detail, or does it mean for whatever reason, much of the time they don't? If I have a 1080p picture with 1080p worth of detail I probably won't downscale it. I'm re-encoding a Bluray at 1080p as we speak, for that very reason.

PS I tried the Ben-Hur screenshot again (720p) and still couldn't retain all the detail in the seated guy's jewellery and clothing, although with a bit of LSFMod() I got it looking much better. The clothing of the guy standing to his left still lost some stripes. Apart from that after downscaling to 720p the rest of the image looked pretty much the same. Would it be enough to stop me downscaling? Maybe..... I'd have to check some more first. If I was only losing detail a small percentage of the time I'd probably still downscale. If it caused noticeable blurring somewhere in the majority of the vidceo I probably wouldn't. Or maybe I'd try 900p instead..... but I'll concede defeat in the case of that Ben-Hur screenshot. My initial impression that'd it'd easily downscale without quality loss was obviously wrong. Although until I played with it again today I was still sure most of the problem was the softer resizing I used the first time. Obviously not....