16:8:8 Color Sampling?

[M Most]

So Sony have invented some new terminology for chroma sub-sampling notation, which is an area of great confusion to begin with.....Now Sony are indeed up to their usual games with their current announcement. Some things in this industry don't ever seem to change.

They haven't invented new terminology, and as someone who works for a company that HAS invented new terminology (and redefined some existing terminology) one shouldn't be throwing such stones.

The 4:2:2 and 4:4:4 designations are based on standard definition video, D1 in particular, and haven't been relevant in an HD world for quite some time. But then again, you already know that. The numbers they are using for the new products are more descriptive in terms of what they actually represent, even if you feel they're unnecessary. Since we have very few details of the coding/recording format for these proposed products - since they haven't actually been publicly discussed yet - I would suggest not slinging arrows, just as Red has asked for restrained arrow slinging prior to their own product announcements. In fact, I would have thought you would be applauding Sony for trying to advance the state of their own art, and by a considerable margin.

 

[Michael Hastings]

It's hard to believe you aren't goofing. Of course... math education sucks so badly in the US, I say as an American, you may not be.

Alex, of course I was goofing - to an extent. I don't want a ratio of gold to silver - I want a specific amount.

I understand ratios, but, as was stated earlier, 4:2:2 and the other ratios also referenced specific numbers and that is the point one presumes Sony is trying to make. This has been a problem throughout the history of video. The fact that SALES people use these numbers to sell against the competition means the MARKETING people have to come up with these things. To an extent every company - including Red - goes a bit beyond the "truth, whole truth, and nothing but the truth..." in their marketing.

As far as too much camera for the money - well of course that was a huge part of RED's marketing - and I'm grateful for that because at $40k to $50k I and probably 5000 or so others wouldn't have purchased a redone.

I don't hold Sony's using their reputation, service and dealer network, etc to get more money for less tech against them. It just means that I won't be buying one because it isn't that important to my particular business model and knowledge/skill base.

But I do wish some of the redusers wouldn't be so strident in their criticism/attacks (and i dont mean you) because to simply deny that those things have value to other businesses that have more capital and different needs just makes them look unrealistic, naive, and/or fanboyish and to some extent reflects on the user base as a whole.

 

[David Mullen ASC]

Unfortunately, I now understand that this terminology isn't something that is set by some third-party committee, but is evidently up-for-grabs to be used and altered by manufacturers as they see fit.

Manufacturers can try to alter terminology and set definitions, but language is fluid and hard to control. Red is no different... last time I checked, they were calling the 30mm x 15mm M-X sensor "Super-35" in size, when Super-35 has traditionally been 24mm wide. Maybe that's changed.

Enough people decide that some word now means "y" instead of "x" and that's what it means, the original definition no longer manners. That's the nature of language, it's a living thing. Manufacturers have tried to get some new terms introduced into the vernacular, or alter old terms, and sometimes they succeed, sometimes not.

We still use the phrase "I filmed something the other day" even when shooting digitally.

 

[Bill Anderson]

Technology is improving so rapidly that the large investment is not what it used to be for a great portion of investors. Just look at the way Jim facilitated the fast transition from RED to Epic purchases: unprecedented. Jim is well positioned, price and quality wise, and his "obsolescence obsolete" model makes more sense every day. One thing is for sure, we won't have to wait a hundred years to see how this plays out.

 

[Michael Olsen]

Two very good points, David. And that was really more where my concern sprang from - with potential codec issues in the future. Where, for instance, is ProRes 161616 for 4K files? Will we ever get that? Or is the difference between a 4K 4:4:4 file and a 4K 16:16:16 file purely semantic? Right now, it's sounding like its semantic, and that 16:16:16 is really just useful when comparing a much larger resolution format to a much smaller format.

...which seems silly to me, because it looks to be all about the ratio of luma to chroma in the YCbCr world. So as said by others, isn't using numbers larger than 4:4:4 (or really 1:1:1) just indicative of the same thing (full luma and spatial subsampling) across different spatial resolutions?

And you are right - manufacturers definitely bend and invent language all the time. I was really mistaken to begin with as to how the language of subsampling was written in the first place.

 

[Alexander Ibrahim]

Hi Alexander, thanks for your reply. I'm beginning to think that, as I move to a RAW world, this YCbCr and other "video" issues is increasingly becoming a non-issue.

With RED, it seems the camera just gives you all there is with a user-selectable degree of compression. Ironically, it's much simpler than the alternative.

--

Also, I really didn't intend on this thread becoming a big thread about Sony. I was really more interested in the terminology. Unfortunately, I now understand that this terminology isn't something that is set by some third-party committee, but is evidently up-for-grabs to be used and altered by manufacturers as they see fit.

Actually, I'm a bit troubled by a lot of the "video" talk.

Since Graeme is on this thread I'd love to hear about it from him.

Here is my take:

Video is optronic (i.e. optical and electronic) imaging technology derived from analog systems design and focused on human perception. By that last part I mean to imply the use of perceptual compression and color subsampling as a technological compromise. This is a common component of such systems, but not a design goal.

Film is an opto-chemical analog sensor technology.

"Digital video" is a video system that converts analog optronic data into digital data.

"Digital cinema" is digital video. Its just built to higher standards. There are fewer compromises. This isn't due to a fundamental difference, but rather to the march of technology. I really think of this term as marketing.

We still have an analog optronic sensor system that converts data to digital.

So, its just as fair to call Red cameras RAW digital video as it is to refer to them as digital cinema.

The closest analog is digital still cameras. We call the $99 "Elph" point and shoot a digital camera. We call a Canon EOS 1DMK4 a digital camera. There is a huge difference of course. Continuing that metaphor, the Canon G12 is the stills equivalent of a Scarlet Fixed. Its still a RAW digital camera, but its a heck of a lot less camera than a 7D or 5DMK2.

The difference is the system design (interchangeable lenses, sensor size and quality, image processing technology, and RAW image recording.)

Bringing it back to the 16:8:8 business. There are truths behind the terminology, but Sony is totally marketspeaking with that particular tidbit. Hence my comments about changing the terminology.

At the end of the day, we have to characterize a lot of technical data to even begin to understand the full picture.

As an example, the terminology I have proposed here would identify an HPX300 recording AVC Intra to P2 internally and an F3 recording to a KiPro Mini identically, when the F3 would be a much higher quality solution, and much more costly. Ferreting out the essential differences requires a lot more data and technical specification.

(Of course the HPX300 has a lot of advantages for certain types of production over F3, Alexa and any proposed Red. That suitability depends on industrial design and ergonomics more than camera technology though.)

As others have said its best to just look at the literal pictures. I use look figuratively- I think running key and color correction tests is a huge part of "looking."

Far too many "testers" treat camera systems like they should treat BluRay- as a delivery system. XDCAM EX 35Mbps can look every bit as good as BluRay h.264, and often better. Of course you can't really create cinema grade results with that source material as the input point for a workflow- but if your ending point is close enough to the sources (like in documentary work) it may not matter at all.

Thought must be given, and tests must be performed, that reflects the full post cycle likely to be used for a system.

I believe that even with the "looking" done well, you still need to characterize the technical systems design in order to really understand the full cycle system performance- acquisition to delivery.

 

[Bill Anderson]

The recent announcements certainly help undermine the equally recent "1080P is enough" cries.

In the face of growing methods of delivery, large sensors/resolution and all that this entails, might prove to be the very thing the large screen needs to breath new life into it, to once again decisively separate the cinema from other means of watching movies.

 

[Alexander Ibrahim]

Two very good points, David. And that was really more where my concern sprang from - with potential codec issues in the future. Where, for instance, is ProRes 161616 for 4K files? Will we ever get that? Or is the difference between a 4K 4:4:4 file and a 4K 16:16:16 file purely semantic? Right now, it's sounding like its semantic, and that 16:16:16 is really just useful when comparing a much larger resolution format to a much smaller format.

ProRes 444 is already 16:16:16... i.e. it does represent 100% chroma resolution.

I don't think it supports 4K yet, and its "only" 12bpc, but that's all that's missing.

As an example, Cineform 4K is already a 12bpc 4K 100% chroma solution. So we already have that.

... 4:4:4 (or really 1:1:1) just indicative of the same thing (full luma and spatial subsampling) across different spatial resolutions?

A quibble... probably a typo on your part.

"(full luma and chroma subsampling) across spatial resolutions."

Then yes.

Of course if you have 100% "subsampling" then you really no longer have SUBsampling ... its just 100% sampling.

 

[Jeremy_Sawatzky]

this whole thread is awfully confusing. i think il just shoot with scarlet and know whats coming out of that camera is the best possible quality at this point.

 

[Gavin Greenwalt]

this whole thread is awfully confusing. i think il just shoot with scarlet and know whats coming out of that camera is the best possible quality at this point.

Welcome to the video world. Where everything is now unnecessarily confusing because of legacy compromises and features which have long gone obsolete.

Pixels should be square. Color should be in RGB (or XYZ). Frame-rates should be integers. Interlacing should be banished. Blacks should be 0 and White should be 1 (And super blacks should be <0 and super-white should be >1) etc...

The video/broadcast world was a hack built on a hack built on a work around delivered in another hack. But it unfortunately still infects what could otherwise be rational and straightforward/simple modern solutions. (I'm looking at you H264)

 

[Alexander Ibrahim]

this whole thread is awfully confusing. i think il just shoot with scarlet and know whats coming out of that camera is the best possible quality at this point.

Haha ...

it will be tempting to think what you wrote is true, and wrong.

Scarlet cameras will shoot 3K, and really give you a nicely oversampled 2k/1080p deliverable.

I think its, in part, RED's answer to the "1080p is fine" crowd, while still maintaining some hope of including your footage in 4K deliverables.

Its worthwhile understanding this stuff, so I hope you'll give the thread another whirl.

It is all a hackish compromise, and I'll be glad when its all history. That won't be for some time though.

 

[Leo Sardello]

I believe that this video might help clear up the matter up a little. It’s certainly how I feel about it anyway.

http://www.youtube.com/watch?v=Y0Z0raWIHXk

 

[Mark Collins]

Technology is improving so rapidly that the large investment is not what it used to be for a great portion of investors. Just look at the way Jim facilitated the fast transition from RED to Epic purchases: unprecedented. Jim is well positioned, price and quality wise, and his "obsolescence obsolete" model makes more sense every day. One thing is for sure, we won't have to wait a hundred years to see how this plays out.

See the point?

Of course you don't. You have to drive 6 hours backwards because you missed it by so much.

The point of this thread is such: Sony has come out with a new sensor that can sample at a high enough rate for the HD and beyond era. So really, 4.2.2 is now 'obsolete" and 16.8.8 is now the forseeable future. However, all of what Sony has said so far is just marketing, and really, until we get our hands on it, is just hype until proven otherwise. The whole debate is whether or not Sony is right or wrong with their math. Not whether or not Red is a good camera.

You know what solves the whole problem: Shoot film.

 

[Graeme Nattress]

But HD is 22:11:11 so what does 16:8:8 actually mean? This is not if Sony have their math right or not - that will come later. It's about what on earth are they on about. And why are they referring to chroma sub-sampling ratios in relation to a sensor rather than to a Y'CbCr signal which is what the notation was meant to describe.

Graeme

 

[Bill Anderson]

Mark, that point seemed so obvious as to make it not worth commenting on: sans camera. Shooting "film" solves no problems for me. My "problems" transcend film ( :

Edit: Graeme says it more effectively

 

[Bill Anderson]

I was thinking, since sensors are getting larger and larger, perhaps we can express aspect ratios thus: 2:1 - 4:2 - 8:4 - and a whopping 16:8. ( :

 

[Stuart English]

Where does 16:8:8 come from ?

Where does 16:8:8 come from ?

Now with a camera that works with a 4K/quad-HD, that's 4X the amount of information, therefore everything else needs to be multiplied by 4, right??? Seems like a logical thing for a marketing executive to assume, right? So now, when you multiply your output resolution by 4, you must (in their minds) also claim your 4:2:2 is really 16:8:8.

That is a good point, if you reference a Quad HD deliverable back to HDCAM SR (the Gold Standard for 1080p broadcast delivery) it does make perfect sense to say 16:8:8 if you are currently saying HDCAM SR is 4:2:2. But then a Quad HD resolution DPX file from REDCINE-X would be 16:16:16...

So like Graeme I am puzzled... I'm sure it will be explained later...

 

[Jeff Kilgroe]

That is a good point, if you reference a Quad HD deliverable back to HDCAM SR (the Gold Standard for 1080p broadcast delivery) it does make perfect sense to say 16:8:8 if you are currently saying HDCAM SR is 4:2:2. But then a Quad HD resolution DPX file from REDCINE-X would be 16:16:16...

I think you missed the sarcasm (sorry for not labeling it, hehe). I don't know if it makes sense to say 16:16:16, or 16:8:8. As we don't really know what they are doing.

So like Graeme I am puzzled... I'm sure it will be explained later...

Here's another theory. Perhaps what's really going on here is they are trying to apply a conventional ratio system to a 16bit RAW paradigm. It's not really a resolution thing at all, but a bit depth thing. Most people in this business think of 4:4:4 as being "full color quality" or something close to that. By Sony saying 16:8:8, they're making a bold marketing statement of "but look, we're a lot bigger and better". In reality, it seems they really may be saying that the chroma components are half of the luma in relation to a 16bit RAW space. We don't know specifics for this sensor. It could be marketing mumbo jumbo... Or it could be that their codec processes in blocks of 16 pixels and aligns chroma info accordingly. Like 4:2:2, but bigger in the sense that it works with a larger Y'UV group. ...But I thought this was RAW? So maybe they are talking about 16bits and relative 8 and 8? wtf? who knows...

So it seems the sensor is 1.85:1 and has a photosite resolution of 8192x2214. But we don't know if it's a checkerboard Bayer pattern or something else, unless I missed something. I did read something some months back, related to Sony, that suggests it may be something like this...

 

[Pietro Impagliazzo]

Some things in this industry don't ever seem to change.

And some do... That's where RED comes in.

 

[Michael Olsen]

Thanks again for all the replies, everyone. I think this is starting to make a little bit more sense to me, with a few mental roadblocks.

It seems like the big issue here is that unknown to myself and probably many more, the calculation that produces a 4:2:2 or 4:2:0 or 3:1:1 ratio takes into account not only the ratio and spatial arrangement of YCbCr, but also the spatial frequency as a component of the equation.

Theoretically, what we would like to see could be known as 1:1:1. It would have equally sized, equally spaced representations of each color. And it could be taken as a spatial resolution independent representation, since it isn't factoring that in at all.

When we add sampling frequency to the mix, 1:1:1 becomes 4:4:4 to represent the spatial resolution of SD 720x480 footage. And only 720x480 footage.

Likewise, using the same sampling frequency idea, 4:4:4 should be represented as 22:22:22 footage in terms of rec.709 HD at 1920x1080.

Extrapolating from that idea, 4K should be ~300:~300:~300. (My ability to calculate the sampling frequency of 4K is limited by my understanding of how all of this functions mathematically).

That all makes sense to me, and I get it. Three thimbles full of water (SD) are not equal to three buckets full of water (HD) which in turn aren't equal to three bathtubs full of water (4K), though the ratio (1:1:1) remains equal.

--

Now where I'm confused, as others seem to be, is the 16:8:8. It would seem, according to this explanation of things, 16:8:8 wouldn't even be the 4:2:2 equivalent in HD. It would be something like an equivalent 2.9:1.5:1.5.

--

It also seems, purely in regard to codec and encoding choices (taking ProRes 444 for a common instance), that the 444 notation is perhaps misapplied...at least based on this understanding, since the codecs appear to encode based on the ratio and are independent of resolution. I.e. a 720x480 ProRes file could actually be 4:4:4. But a 1920x1080p ProRes file that was 4:4:4 would really be relatively significantly undersampled. I trust that this isn't happening, and that instead, 444 has become the terminology to use, regardless of resolution, for "1:1:1" or Full:Full:Full. As such, I believe encoding 1080p with ProRes 4:4:4 would result in a spatially-relevant 22:22:22 file, providing that color fidelity was there to begin with in the source.

In other words, working from a nice .r3d and using ProRes444, I won't be missing out on a whole mess of luma or chroma data that would be encoded by a non-existant ProRes222222 or ProRes3003000300. Or perhaps another way of stating it, codecs don't care about spatial resolution - they will reproduce the color ratio at any supported resolution.

...is any of this sounding right?