Steve Yedlin, ASC on Resolution

[Michael Lindsay]

Thanks Joseph for answering..

It depends on the type of shooting you are doing.
For the most part if you are doing camera tests and properly lighting for the look you are expecting, most high level professional cameras are interchangeable for the bulk of work, even into HDR.
It comes down to operation preference and workflow for the most part.

Basically any camera that can't capture an actor with accurate skintones in a few stops above and a few below will not be on the market for long. In that sense cameras have to be similar.

Steve Yedlins display prep demo show this pretty well. For the most part you can make all of the professional cameras look the same and he does a good job giving a number of different cameras a consistent look. It's exciting from my perspective because post houses do that day in and day out, but can never show the concept in action because we do it for clients on their footage.

 

[Jamison Stark]

As a metaphor for people that think its pointless to shoot higher then hd/4k, I like to use "shrink plastic" (not sure its the right term in English) As a kid we did alot of them. You paint something on the plastic and when its shrinks in the Owen, and in the process everything gets more smooth and crisp. Your artifact from coloring your artwork disappears. Could not say the same thing when stretching something.

Shrinky Dinks! ;-)

 

[Stacey Spears]

Good to see that these filters are still being researched and new ideas tested. But the differences are now marginal between the best old filters and new ones unless one zooms in quite some.

I will process a Catmull-Rom and Lanczos to show the difference. Both of those have a lot of ringing in them. True, you may need to pixel peep to truly appreciate the difference. I will sleep better at night knowing I extracted every last pixel from my images.

Mathias has done some great work on upscaling too. However, I could not agree more with Graeme, the best scaling is down. In both cases, scaling improvements have been stagnate for years.

I really don't like the studios upscaling movies to UHD resolution for UHD (streaming, BD, etc...). Let the display scale and release the movies at their native resolution and take advantage of HDR and higher bitrates. I will be showing 100 Mbps HD, 100 Mbps UHD and 15 Mbps UHD on my next disc. I know one OTT that thinks 15 Mbps is too high for UHD and they plan to lower the bitrate. I hope my bitrate demo makes people demand for more bits, not less.

Sorry for going off topic.

 

[Stacey Spears]

Shrinky Dinks! ;-)

Fond memories.

 

[Matt Bai]

Also, for all the Monday morning Quarterbacks in this thread.... Steve Yedlin shot Star Wars episode 8.

Remind me which episode you're shooting?

What?

 

[Marc Wielage]

I know one OTT that thinks 15 Mbps is too high for UHD and they plan to lower the bitrate. I hope my bitrate demo makes people demand for more bits, not less.

I want a T-shirt that says "I Demand More Bits!"

 

[Michel Hafner]

I really don't like the studios upscaling movies to UHD resolution for UHD (streaming, BD, etc...). Let the display scale and release the movies at their native resolution and take advantage of HDR and higher bitrates.

You can't release 1080p4:4:4 on BD. The only way is upscaling to UHD 4:2:0 for UHD BD. I prefer that to BD's 1080p4:2:0 upscaled by some display.

 

[Gavin Greenwalt]

So the idea was, what if I only have 2K of this 6K image — what if I have to choke it all the way down to this. What does that image look like in comparison to the 6K original after it is scaled up to 4K? And then we see how very little the resolution actually matters; it’s almost indistinguishable. And from an audience perspective, it’s no big difference in terms of their perceptual experience. What you can see, however, is that while you get almost no difference from the resolution change, you do get a difference from the scaling algorithms used to go from 6K to 4K and 2K to 4K.

It only is indistinguishable because he started from a 6k capture. Yeah, a 2k master from a 6k capture looks fantastic even scaled up to 4k. But the 2k only looks as good as it does because of the 6k master. This is true of CG. Render something with 1x primary ray sampling and it looks like garbage. You usually end up rendering at many dozens of samples per pixel even for 2k rendering.

The real test isn't to take a 6k image and the scale it down to 2k and then back up to 4k it's to take a 6k camera shoot it wide so that you're only using 2k worth of sensor pixels and then scale it up to 4k. You'll see a huge difference. In fact it's pretty obvious to me that you can see in the Alexa XT vs 65. This image has been scaled down to 25% of its size in the video. It's been run through MP4 compression, it's been resaved as a JPEG and it's still substantially sharper with the 6k camera than the 3.2k camera. He provided the perfect counter argument to his argument in the video while talking about how resolution of the camera doesn't matter.

Ultimately if your argument is "the audience won't notice" then you could pretty much quit with SD and not have cinematographers at all. There is a lot of things that we do for the art that 99.9% of the audience just doesn't care about. I think it's safe to say that at most 1-2% of the audience can tell the difference even between a spherical and anamorphic lens let alone different spherical lenses. And undoubtedly there are diminishing returns on resolution so I think it's fundamentally a fair point. But this test is not a fair way to demonstrate it since it since it cheats.

I agree his arguments about "K isn't everything", (compression etc do have a huge effect) is a very true one (and I've complained often about RED's debayer being comparably very soft until the latest update) but this very true argument is let down by his repeated assertion that Ks have no intrinsic value. Also some of his concerns such as "providers will sacrifice compression to achieve resolution" are reasonable but not born out by reality. It's true that there is often an overlap between compression bitrate and resolution. But it's an overlap and good providers like Netflix are aware of and pick which wholistically to deliver the best quality at any bitrate. https://medium.com/netflix-techblog/per-title-encode-optimization-7e99442b62a2

Also some very erroneous comments like "We don't need more resolution if we have halation 30 pixels across". That's just nonsense. A lens flare is a form of halation. If you point a light into a lens you'll get halation 6,000 pixels across. But in most shots you don't have that extreme of a contrast situation where something is so bright that the halation is *visible*. Just because the contrast between the sun and a sillouette results in erosion means that you'll get that degree of erosion on an image with a regular 3-4 stops of contrast between pixels.

He's very obviously picked a conclusion and then is just looking for reasons that support his conclusion anywhere it can be found. The real takeaway from this video is that Arri did a phenomenal job on the Alexa65.

 

[Gavin Greenwalt]

I just stuck a camera out the window. If the average viewer can't tell (from only 1080p downscales!) which was shot at 6k and which was shot at 3k I'll eat a comically large cowboy hat. (and the 3k image also was sharpened for extra good measure).

(Workflow: Shot one at 6k, Shot one at 3k, resized both to 1080p and extracted a 500x500 px 1:1 crop of the downsampled image.)

 

[PatrickFaith]

..
The real test isn't to take a 6k image and the scale it down to 2k and then back up to 4k it's to take a 6k camera shoot it wide so that you're only using 2k worth of sensor pixels and then scale it up to 4k. You'll see a huge difference. In fact it's pretty obvious to me that you can see in the Alexa XT vs 65. This image has been scaled down to 25% of its size in the video. It's been run through MP4 compression, it's been resaved as a JPEG and it's still substantially sharper with the 6k camera than the 3.2k camera. He provided the perfect counter argument to his argument in the video while talking about how resolution of the camera doesn't matter.
..

Ultimately if your argument is "the audience won't notice" then you could pretty much quit with SD and not have cinematographers at all. There is a lot of things that we do for the art that 99.9% of the audience just doesn't care about. I think it's safe to say that at most 1-2% of the audience can tell the difference even between a spherical and anamorphic lens let alone different spherical lenses. And undoubtedly there are diminishing returns on resolution so I think it's fundamentally a fair point. But this test is not a fair way to demonstrate it since it since it cheats.

...

Also some very erroneous comments like "We don't need more resolution if we have halation 30 pixels across". That's just nonsense. A lens flare is a form of halation. If you point a light into a lens you'll get halation 6,000 pixels across. But in most shots you don't have that extreme of a contrast situation where something is so bright that the halation is *visible*. Just because the contrast between the sun and a sillouette results in erosion means that you'll get that degree of erosion on an image with a regular 3-4 stops of contrast between pixels.
...

He's very obviously picked a conclusion and then is just looking for reasons that support his conclusion anywhere it can be found. The real takeaway from this video is that Arri did a phenomenal job on the Alexa65.

I read the ASC magazine when it comes out, I think the top tier guys get it but the entire middle seems to be basically stupid - I mean they somehow got their because all the technical work was done in the past and they just had to copy things, and they don't have the capacity to learn new tech. If I was red this would drive me insane.

 

[Graeme Nattress]

Gavin, that's a great example. It's even better than mine as mine had the advantage of starting with a higher resolution to begin with.

Graeme

 

[Mark K.]

Ultimately if your argument is "the audience won't notice" then you could pretty much quit with SD and not have cinematographers at all. There is a lot of things that we do for the art that 99.9% of the audience just doesn't care about.

I think that's rather an oversimplification. Certainly 'more' is always 'better' objectively. But there is a point of diminishing returns, not to mention the fact that we shoot motion pictures at 24 fps with a 180 degree shutter, which renders the bulk of our resolution (in perhaps the majority of the frames we shoot) a 'k'-less blur anyway.

And we're also talking about an essentially fixed medium - video footage displayed or projected on set screen sizes (from handheld smart phones, to flat screen TVs, to 50-60' cinema screens) and all viewed from relatively standardised distances.

Now are there situations where high-resolutions are necessary? Absolutely. Anything you're putting on an IMAX screen (or anything with a comparable display:viewing distance ratio) needs more. The point is simply that such viewing conditions are a statistical outlier in our viewing habits. And until the way we watch things changes, the size of the screens we use increases significantly, or our viewing distances from those screen reduce considerably - there's no real need to change things. As we're already at the point of diminishing returns.

 

[Stacey Spears]

You can't release 1080p4:4:4 on BD. The only way is upscaling to UHD 4:2:0 for UHD BD. I prefer that to BD's 1080p4:2:0 upscaled by some display.

Kris Deering and Darin Perrigo had suggested that pre-scaling 1080p to UHD would result in more chroma resolution about a year ago, so I did some tests and found this to be less important than a higher bitrate. People picked the higher bitrate 1080p over the pre scaled UHD. Both were viewed at UHD in the end. It was an interesting idea, but just goes to show why we get away with 4:2:0 to begin with. On a test pattern, much different story.

The artifacts from scaling up, such as ringing, increased noise size and the sharpening applied to compensate for the softness from scaling up all made compression more difficult, requiring an even higher bitrate.

If your HD masters are 10-bit 4:2:2, there is no 4:4:4 to begin with, which is real for some.

Dolby Vision, on UHD BD, is still 4:2:0 but they get two more bits over HDR10, which gives you 12-bit. Again, they did actual testing and figured out which provided real visual improvements (12-bit) and which did not (4:2:2)

I think it was a clever idea to retain more chroma resolution, but does not pay off in the end.

 

[Michel Hafner]

Kris Deering and Darin Perrigo had suggested that pre-scaling 1080p to UHD would result in more chroma resolution about a year ago, so I did some tests and found this to be less important than a higher bitrate. People picked the higher bitrate 1080p over the pre scaled UHD. Both were viewed at UHD in the end. It was an interesting idea, but just goes to show why we get away with 4:2:0 to begin with. On a test pattern, much different story.

The artifacts from scaling up, such as ringing, increased noise size and the sharpening applied to compensate for the softness from scaling up all made compression more difficult, requiring an even higher bitrate.

If your HD masters are 10-bit 4:2:2, there is no 4:4:4 to begin with, which is real for some.

Dolby Vision, on UHD BD, is still 4:2:0 but they get two more bits over HDR10, which gives you 12-bit. Again, they did actual testing and figured out which provided real visual improvements (12-bit) and which did not (4:2:2)

I think it was a clever idea to retain more chroma resolution, but does not pay off in the end.

It depends. The relevant sources for upscaling are usually 2K DI files with 4:4:4 and not some 1080p master. So first is the issue of cropping and upscaling or upscaling with a fractional factor. The former should look better but loses some image. Then the scaling algorithm is important. And the bit rate it is compressed with at the end. If all is optimised the result should look superb and better than some display upscaling 1080p with off the shelf chips. And the chroma resolution difference is very visible for some images (and not much for others). The best examples are coloured credits with saturated hues. They go from blurry to sharp (Blade Runner, Blade...).

 

[Blair S. Paulsen]

All other things being equal, mo' resolution is mo' bettah.
Compression is not fundamentally bad, over compression is.
Program delivery modalities need significant improvements.
Oh, and anyone who sits too far back from their display should be arrested ;-)

Cheers - #19

 

[Joel Arvidsson]

Anyone who sits too far back from their display should be arrested ;-)
Cheers - #19

I down with that!

 

[Fabricio Morato]

Steve's video made me want to try to do this for myself. I'm also using nuke and since he uses some group nodes, and never shows any config, I have no idea what are the scaling algorithms he is using. I've tried to replicate his upscaling methodology but so far it only points to the opposite of what he's showing (that the difference is minimal on 1:1 viewing).
I've used all of nuke's built in scaling algorithms and so far I'm not getting even close to his results.
Any idea what he is doing? Also about idea about his "clamp" group?
I've started from a single 100MP image and downscaled (using all algorithms available) to 2k, 4k, 8k and upscaled to 4k, 8k, 11.4k and 16k. There are HUGE diferences for me here. Only on the 16k upscale (viewing at 1:1) that the 8k and 11k (original) have "almost" no visible diference. But for the rest it's showing big differences.

Any pointers would be appreciated, since I would love to have the same kind of results as his. Thanks

edit: btw the best scaling algorithm I've found to be (inside nuke) was TVI, but even this one does not give the same kind of result.

 

[Jeffery Anderson]

Are audiences complaining?

 

[Blair S. Paulsen]

Are audiences complaining?

This fallacy makes me CRAZY. Unless you expose viewers to side by side (or sequential) comparisons they don't know what they don't see.

IMO, this same disconnect (pun aspirational) is why cable/satellite MSOs have chosen to stomp on (over compress) bit rates per channel to open up space for niche channels few watch. Huh? Imagine an audience survey (or complaint log) where most of the feedback is people requesting specific channels they have high interest in. Sure, there are a few complaints about picture quality, but how many?

The program producers, networks, distribution infrastructure companies, etc are unlikely to increase their costs if it's not going to increase revenues. OTT services like Netflix are motivated to market any differentiator, like resolution, to make inroads against established players - so the dam won't hold forever. That said, data caps and the fact that the same entity that sells you TV service is also likely to be the same one who sells you bandwidth for IP delivered content mitigates the amount of leverage new media has to actually move the IQ needle.

Bottom line, virtually all perception is "compared to what". I've had the privilege, and curse, of seeing uncompressed imagery in calibrated color suites - then seeing the same material at home, beaten down to mush. IMHO, even to a less sensitive audience, an A/B butterfly of those two versions of the same content would be a jaw dropper.

Cheers - #19

 

[Marc Wielage]

This fallacy makes me CRAZY. Unless you expose viewers to side by side (or sequential) comparisons they don't know what they don't see.

We run into this kind of thinking all the time, where clients want to let the little things go in final color, the final mix, and in the final master. I'm always quick to remind them, "the problem is, the more little things we let go, the more risk we have that it'll all start to snowball and ultimately look bad." There's also a point where the audience may not know exactly what's wrong with the picture, but they'll feel uncomfortable with it on some level. And we don't want that.

I cringe whenever I'm working on a "pure" 4K show and then they slug in some wonky HD up-rez stock footage that has ringing and nasty jagged edges all over the place, just a really crappy upconverted H.264 file or something. Hate it.