Consumer Technology Association sets official standard for 8K UHD TVs

[Sttratos]

For filmmakers, 4K has already firmly been the standard.

What filmmakers? Like Roger Deakins and the people he shoots for? Who shot Alexa 2.8 and 3.2 without giving half a thought to all the other higher resolution cameras already out at the time? Or any of the hundreds of filmmakers who did the same and shot Alexa?

And 4K is still not any standard. For how long have movies been shot and finished in 4K now? It is becoming a standard. Solid? Not yet. There is still stuff being finished in 2K all over.


Besides, filmmaking is such as small slice of the pie anyway.

 

[Sttratos]

Dan, you know I respect you, but every time I see something like this, to me it reads like someone is saying, "Why bother... With anything?"

Yes, people are consuming tons of content on their cell phones and tablets, myself included, BUT there are also lots of people consuming content on large, high resolution screens, too, myself included(92" 1080 and 55" 4K HDR and may go 120" 4K HDR front projection in the next few months to replace the 92"). I mean, way more people drive "normal" cars than jacked-up 4x4's with monster tires, but they still make and sale the lift kits and monster tires.

Yes, but the jacked-up 4x4's are not the standard and wouldn't be called such by any sane person. So wake me up when the zombie apocalypse is upon us and jacked-up 4x4's are actually necessary in daily drive. As opposed to some fashion statement or trial at looking cool.

 

[Sttratos]

Some very thoughtful replies and I'm really happy a lot of you "get it" with all of this. It was not very long ago where I was doing a panel on 4K at NAB and discussing the emerging trend in the face of a good amount of people who thought 4K would never catch on. Fast forward to present day, every camera manufacturer is now making 4K or 4K+ solutions and 4K televisions are cheap.




You'll have to decide when standard/normal is and when you want to wake up I guess. I personally haven't delivered under 4K this decade, but I also sort of specialize in the 4K, 8K, and beyond arena. I know my market isn't everybody's market.

To your point, 4K is still not the "mass market" globally, but what you should take note of are:
- At the end of 2019 in the USA UHD 4K televisions hit about the 50% saturation point in homes. Basically around 1/2 of every home in the US will have a UHD 4K TV, give or take a few percent, we don't know how holiday season is playing out.
- There are about 300+ million homes with UHD 4K televisions today, not including businesses or venues. Also not including computer monitors and laptops.
- UHD 4K sales projections from 2017-2019 have all been exceeded by notable margins each year. Many expected a fairly linear growth in this time frame, it's been more of a steep ramp in the last two years in particular.
- Last year's projections put 2023-2025 as the moment when we'll have about 1 billion UHD 4K televisions out there worldwide. I personally think we'll get there in 2022-2023. Especially once we see how things play out in 2020.
- Nearly every current streaming service, the ones launching in fall, as well as the ones through April 2020 and likely beyond are all producing 4K content.

There was a time when "nobody had an HD TV". I don't think many of you are replying to me from sub-HD devices

On a personal level, I'm just happy UHD 4K is getting so much traction and adoption. Also happy to see UHD 4K TVs are extremely affordable these days. 8K is still very early on, but I assure you it's happening. This is all really at the moment about deep 4K market penetration and emerging 8K, which will ramp up in the coming years. That actually has been my main market for a bit now and that 8K content also gets finished out at 4K as well.


The vast library of SD and HD content getting up to snuff for these higher resolution displays is a fun journey to watch right now. Ideally every film ever made could go through the tedious process of remastering, which is expensive and often not possible due to degradation and loss of assets. However, upscalers in TVs are getting better and smarter. There's also some advanced AI processing being deployed by 3 main companies. Pretty impressive results. You still miss out on the fine details, like subtle texture, but it's an improvement to a simple upscale for sure.

But really this is all about current and future content doing it "right" with proper 4K or 8K finishes.

I thought long and hard if I should reply to this. Because, what other position can one expect from a Red advocate really? Red sells resolution. It's a bit like expecting a catholic cleric to say and stand by anything besides what's on the bible. So a bit of a moot point getting in any sort of back and forth with you.

But just a couple of points:

You'll have to decide when standard/normal is and when you want to wake up I guess.

Well frankly, I or you don't get to decide anything. The WORLD does. 4K will be standard when, and I'm talking world or at least western world, not only the U.S. or LA:

-All or the huge majority of video content is delivered in 4K , across ALL platforms and viewing devices.
-All or the huge majority of broadcasters are broadcasting in real 4K or 4K at all.
-All or the huge majority of movies are shot and finished in 4K. The exception being selected shots or movies shooting something else for an effect. No different than shooting 16 or even S8 for artistic reasons.
-All delivered 4K content can be viewed as 4K, by everyone or the huge majority of the world. At least western world. There were some western broadcasters still using SD cameras and bumping to HD for broadcast till just a few years ago. We are only now getting to the point of real and solid HD broadcast actually.
-All 4K streaming can sustain that resolution while displaying and all people have access to that level of sustainability. Without 4K getting bumped down to HD or even worse because streaming can't keep it up. 4K streaming at the moment is almost false advertisement in most markets. Producing 4K content is not enough. Having a 4K TV is not enough. What's the point of watching 4K content on a 4K TV if you are not seeing 4K?
-All or the huge majority of people will bark at watching HD content because they think the quality is so low. As it happens when people watch SD or something VHS originated for example now.
-All or the huge majority of people throw away their HD Blurays in a similar way they did they VHS tapes. Then adopt 4K Bluray. Not just video buffs but the huge majority.

I could continue. But I guess I made my point. All the above won't just happen "overnight." Regardless of how optimistic everyone selling the tech tries to come across. If it does, I will gladly stand corrected. And I never said it is not coming.

I personally haven't delivered under 4K this decade, but I also sort of specialize in the 4K, 8K, and beyond arena. I know my market isn't everybody's market.

Exactly. You said it yourself. You live in a bubble. I know so many shooters who are never requested 4K. There is more to the production world than LA, as hard to believe as that may seem to be for some there.

Anyways, I shouldn't have to explain what standard/normal means. I think it's pretty obvious it means something everyone or the huge majority uses, does, agrees etc. SD was a de-facto standard. We had PAL, SECAM and NTSC and color and luma details. But it was a standard. Because it was around long enough to become widely adopted across the board. HD barely made it in comparison. 4K is in no way there and to be honest I think it never will. It will probably not have the time. 4K is the new 720p. 8K will be here before 4K truly becomes any standard. But yes, wake me up then.

 

[Run&Gun]

Yes, but the jacked-up 4x4's are not the standard and wouldn't be called such by any sane person. So wake me up when the zombie apocalypse is upon us and jacked-up 4x4's are actually necessary in daily drive. As opposed to some fashion statement or trial at looking cool.

That was kind of my point. It may not be a standard, but there is a large enough base to warrant the manufacture and sale of. I wasn’t trying to draw an exact parallel. 8K(screens) will definitely become more main stream and standard than a jacked up 4x4.

 

[Cary Knoop]

One small issue with a RGB sensor is we would suddenly have raw data or lightly compressed video 3x larger than bayer raw video. Bayer definitely has its faults but it really acts as a great form of image compression where we can get a RGB 4:4:4 image out of 1/3 of the data. Formats like true uncompressed and ProRes4444 are now a total waste compared to raw formats like Braw and ProRes Raw. Visually any uncompressed or lightly compressed video is from a bayer sensor anyway so they gain absolutely nothing at all except a much larger file.

Bayers sensors will always have that unique ability to deliver what most consider to be an amazing RGB image with 1/3 of the data. Even in the stills world that do pixel peep and crop to extreme levels there is rarely a complaint about bayer sensors.

I would love to see 3 chip cameras make a come back but recording a 444 format with something like that is going to be heavy.

For effective compression, there is absolutely no need to resort to Bayer patterns, chroma subsampling, YCbCr color encodings, instead, it can all be handled, more effectively, by just compressing RGB 4:4:4 signals.

 

[Cary Knoop]

For how long have movies been shot and finished in 4K now? It is becoming a standard. Solid? Not yet. There is still stuff being finished in 2K all over.

It may be finished in 2k now but one would hope we don't have filmmakers that are so short-sighted to only record for present-day needs.

 

[Thomas Smet]

It may be finished in 2k now but one would hope we don't have filmmakers that are so short-sighted to only record for present-day needs.

Most of the Marvel movies were shot on Alexa at 2.8k. Sometimes they would use open gate at 3.2k These are some pretty big budget movies full of VFX that likely have earned more money and sold more copies than most 4k shot material put together. The more recent movies were shot at higher resolution but honestly it doesn't make the movies better or worse. Guardians of the Galaxy is one of my all time favorites and it is 2.8k and will continue to be one of my favorites for many years to come when everybody is talking about how only real movies are shot 4k and 8k. Not saying if a camera is as good it should just go ahead and use 4k but that in fact the Alexa is used on the majority of Hollywood films and it is still lat the end of the day a 2.8k camera.

 

[Thomas Smet]

For effective compression, there is absolutely no need to resort to Bayer patterns, chroma subsampling, YCbCr color encodings, instead, it can all be handled, more effectively, by just compressing RGB 4:4:4 signals.

Yes if one wants higher image compression. A 3:1 compression is still going to be 3x higher than it is with a bayer sensor. A bayer sensor is only one channel worth of data. RGB is 3. It will always be 3x the data no matter what.

I'm also sure you can't actually demonstrate any realistic flaw in using a bayer sensor. I'm not talking the flaws on paper but any real world flaws. The entire Hollywood world has moved to bayer sensors and like it or not they create very beautiful images that even the most critical film maker rarely finds an issue with. It seems to be only the video nerds that are not film makers that seem to have an issue with it from an engineering perspective.

 

[Cary Knoop]

Yes if one wants higher image compression. A 3:1 compression is still going to be 3x higher than it is with a bayer sensor. A bayer sensor is only one channel worth of data. RGB is 3. It will always be 3x the data no matter what.

But the Bayer pattern is also a kind of compression. In the end a full 4:4:4 compression, not based on any Bayer pattern will always be more efficient.

There are basically two kinds of compression, fixed and variable. Fixed compression are things like Bayer patterns, chroma subsampling, interlacing, fixed bit gamma curves etc.
Variable compression is a combination of lossless and lossy elements where the lossy elements are chosen based on visual human perception.

The best strategy is to take a full 4:4:4 pattern and compress that using floating-point code values without any other fixed mumbo jumbo shortcuts.

Ten years from now we are going to laugh at all this debayer processing that you seem to want to defend so vigorously.

 

[DLD]

FWIW, Deakins preferred 2.8 MPX Alexa ST/XT to Alexa65 for BR2049 because he likes the s35 depth of field and the XT "noise texture" (which, I assume, brought it closer to film) but Nolan and Tarantino preferred the DoF of a 70mm format (though they shot on film while Deakins prefers the workflow of digital).

Of course, that was before FX9 was announced. I am sure all three would switch.

 

[Greg_E]

But the Bayer pattern is also a kind of compression. In the end a full 4:4:4 compression, not based on any Bayer pattern will always be more efficient.

There are basically two kinds of compression, fixed and variable. Fixed compression are things like Bayer patterns, chroma subsampling, interlacing, fixed bit gamma curves etc.
Variable compression is a combination of lossless and lossy elements where the lossy elements are chosen based on visual human perception.

The best strategy is to take a full 4:4:4 pattern and compress that using floating-point code values without any other fixed mumbo jumbo shortcuts.

Ten years from now we are going to laugh at all this debayer processing that you seem to want to defend so vigorously.

I think people missed it when you said the Bayer pattern is a form of 4:2:2 compression. For every 4 green pixels, there are 2 red and 2 blue pixels. I guess that means we just can not get real 4:4:4 out of current single chip cameras unless they are super sampled from a larger pixel count (like 5.7k to get 4k files), and even then I'm not completely convinced this is true as I haven't tested it myself. Seems you should need to take each quad of Bayer pattern as a single pixel to get real 4:4:4 pixels out of an imager, but I'm probably wrong here.

 

[Thomas Smet]

I think people missed it when you said the Bayer pattern is a form of 4:2:2 compression. For every 4 green pixels, there are 2 red and 2 blue pixels. I guess that means we just can not get real 4:4:4 out of current single chip cameras unless they are super sampled from a larger pixel count (like 5.7k to get 4k files), and even then I'm not completely convinced this is true as I haven't tested it myself. Seems you should need to take each quad of Bayer pattern as a single pixel to get real 4:4:4 pixels out of an imager, but I'm probably wrong here.

It is a lot more complicated than that due to how the pixels are arranged in a pattern. With 4:2:2 and 4:2:0 you skip those pixels every line no matter what and you eliminate the U and the V pixels. Bayer alternates every other line so one time the red may be missing but the next line down its the blue missing. Not both the red and blue at the same time. Plus they are offset from each other and when it comes time to rebuild the image each channel only has to fill in the gaps and it has a better arrangement of samples of help create a new averaged pixel from.

It is due to this arrangement why we can get a RGB image from a bayer sensor. Every single pixel has a unique red, green and blue value. Every pixel is interpolated based on an average of surrounding samples which is where the softness comes from but every pixel has all three values.

4:2:2 and 4:2:0 just takes that interpolated RGB image and tosses away the U and V data entirely which is any extra layer of tossing out samples. Unlike bayer it doesn't alternate every other pixel so its much harder to interpolate those missing gaps. Not to mention you are interpolating the samples twice.

So you very much can and do have 4:4:4 RGB from a bayer sensor. Its softer and interpolated but its still RGB.

 

[Thomas Smet]

But the Bayer pattern is also a kind of compression. In the end a full 4:4:4 compression, not based on any Bayer pattern will always be more efficient.

There are basically two kinds of compression, fixed and variable. Fixed compression are things like Bayer patterns, chroma subsampling, interlacing, fixed bit gamma curves etc.
Variable compression is a combination of lossless and lossy elements where the lossy elements are chosen based on visual human perception.

The best strategy is to take a full 4:4:4 pattern and compress that using floating-point code values without any other fixed mumbo jumbo shortcuts.

Ten years from now we are going to laugh at all this debayer processing that you seem to want to defend so vigorously.

Not defending bayer and I know exactly what its limitations are. I just also know the alternative is only theoretical since we don't have 3 chip 4 cameras capable of recording 3 channels of raw data.

I also know a lot of talented and creative people way above our pay grades have no problem with it at all and I'm going to assume they know what they are doing.

I studied and worked in VFX for years and bayer just isn't a huge concern as much as other factors are. Part of the way around that is to just shoot higher resolution like 4k, 5k or 6k and down convert. 4k bayer down to 2k is about as good as a 2k 3 channel sensor.

I also don't think it is accurate to call it a RGB sensor. We use raw not just because its a bayer sensor and it gives us 4:4:4 but because it has no image processing. RGB is the opposite of raw and is a set of data processed into a image we can see. I don't think many in VFX would want to give up the ability to alter the image in post beyond what a baked in RGB image would be. The beauty of higher resolution bayer is we don't have a baked in image and we can process to whatever color space we want and down convert to a virtually perfect RGB image for mastering.

We already have 5k sensors for a 2.5k master. We already have 6k and 8k sensors for a 3k and 4k master. Like you said it will be years before we have a 3 channel solution and by the time we do we may have already moved on from 8k. Bayer will now always be one step ahead. If somebody manages a 3 channel sensor that doesn't suck at 4k bayer will be using 8k as the norm and ready to move to 16k.

The small amount of benefits from such a move are not worth sacrificing sensitivity, dynamic range, resolution and so forth. If somebody can make such a sensor they will be rich but alas many have tried and failed so far.

I wish I knew why we can't have a 3 sensor 4k design? If we could do it in the 90's I'm sure somebody could do it again with better technology and more precise engineering. Sure it would drive up the cost but I'm sure many would be willing to pay extra for that. Perhaps the other reason is it just wouldn't have enough visual impact anymore so nobody wants to bother with it. 3 chip made a lot more sense back when we only had SD cameras and a sensor with even less resolution than 720x480 was pretty darn bad. HD we could still see some advantages to a 3 chip design but even then there were many great bayer sensor HD cameras that were turning heads already. Once we moved to DSLRs and 4k or larger sensors the need for 3 chips was kind of over really. I'm just not sure how the added cost and complexity would be worth it visually to actually make a difference. The need to do so with 8k is even more insane and no human on the planet would tell the difference. There would be zero visual loss from a bayer 8k sensor except for the most extreme pixel peeping.

 

[combatentropy]

Why Bayer, by Graeme Nattress, RED engineer, 4/21/2008

 

[Cary Knoop]

We use raw not just because its a bayer sensor and it gives us 4:4:4 but because it has no image processing. RGB is the opposite of raw and is a set of data processed into a image we can see. I don't think many in VFX would want to give up the ability to alter the image in post beyond what a baked in RGB image would be.

I don't agree with your argument at all. There is absolutely nothing processed about 4:4:4 from a sensor that would capture RGB without any Bayer pattern.
Obviously scene linear with floating-point code values would be the best.

The beauty of higher resolution bayer is we don't have a baked in image and we can process to whatever color space we want and down convert to a virtually perfect RGB image for mastering.

Again it is a non-argument, a sensor not using a Bayer pattern would not have to bake in anything.

It begs the question of what you understand backing-in actually means.

To get video from a Bayer sensor you have to interpolate values from different frequencies, doing that requires you to decide on a colorspace and white balance, that's why transmitting raw data is an advantage because you can defer that until post. But that is not a feature, it is a kludge. A sensor giving RGB would not have that issue, assuming of course it has proper bit-depth in the code values. Again, the best solution would be scene-linear with floating-point code values.

There is absolutely no advantage to a Bayer sensor, but of course, you are right that there are right now no good alternatives. But once there is an alternative we should jump on it, sticking to it would be as backward as sticking to interlaced footage or chroma subsampling and fixed bit code values.

 

[Guest]

Why Bayer, by Graeme Nattress, RED engineer, 4/21/2008

Blast from the past!

 

[Guest]

There is absolutely no advantage to a Bayer sensor

The advantages are pretty clear.

 

[Guest]

I also don't think it is accurate to call it a RGB sensor. We use raw not just because its a bayer sensor and it gives us 4:4:4 but because it has no image processing. RGB is the opposite of raw and is a set of data processed into a image we can see. I don't think many in VFX would want to give up the ability to alter the image in post beyond what a baked in RGB image would be. The beauty of higher resolution bayer is we don't have a baked in image and we can process to whatever color space we want and down convert to a virtually perfect RGB image for mastering.

This sounds a little confusing.

I wish I knew why we can't have a 3 sensor 4k design?

Graeme semi answer this in his 2008 post. Something to do with pixels not perfectly aligning, so the “pixel perfect chroma sampling” concept is voided.

I’d also wonder what the cooling would be like for 3 s35 sensors.

 

[Greg_E]

The advantages are pretty clear.

Yes, the advantage is that we can make Bayer and Fuji patterns, we can not make a single plane RGB sensor (at this moment) hat works well enough and cheaply enough to do this. If we could make them easily and cheaply, Foveon would be in every camera by now.

What has happened to the triangle filter arrays that have been tried? Why are those a failure.

With ever increasing die size, I'm kind of surprised we haven't gone to a triangle stuffed filter array with a micro lens over each triangle to get true RGB of a lower resolution. Should give really clean images and i the micro lens is designed well, can give some help to the wide angle non-retrofocus lenses.

And then there is this https://www.newsshooter.com/2017/09/13/sony-go-organic-with-a-new-vertical-colour-separation-sensor/

 

[Guest]

Yes, the advantage is that we can make Bayer and Fuji patterns, we can not make a single plane RGB sensor (at this moment) hat works well enough and cheaply enough to do this. If we could make them easily and cheaply, Foveon would be in every camera by now.

What has happened to the triangle filter arrays that have been tried? Why are those a failure.

With ever increasing die size, I'm kind of surprised we haven't gone to a triangle stuffed filter array with a micro lens over each triangle to get true RGB of a lower resolution. Should give really clean images and i the micro lens is designed well, can give some help to the wide angle non-retrofocus lenses. Presumably all at the lowest cost. But personally dislike Bayer at the end of the day.

And then there is this https://www.newsshooter.com/2017/09/13/sony-go-organic-with-a-new-vertical-colour-separation-sensor/

Bayer pattern is a compromise. It is not the ideal.

We all wish there was something better available.

That said, with current tech it seems to offer the fewest compromises. Allows one sensor to maximize the resolution and still give good results in color.

I personally dislike the bayer pattern and similar filter designs. Unless, there is a need to maximize luma resolution.