4:4:4 ?? [Archive] - REDUSER.net

purefilm

10-04-2007, 06:10 AM

Thank you for explaining that. I don't think many people understand this, because I was talking to some professionals at a rental house about the camera and they asked me is it 4:4:4, and I tell them no they turned there nose at me with the notion that the camera was inferior in some way. I tell them that I think what the red does is better than 4:4:4 but that I have no idea how to explain this to them.

Graeme Nattress

10-04-2007, 06:16 AM

I'd just say, to keep it simple - "No, RED shoot's RAW". And if they ask what RAW means - reply with: "Well, you've shot RAW on your DLSR, right? No? Ok, it means you can non-destructively change exposure, white balance, gain, and how the image looks in post, without loss. It means it's hard to screw-up in camera. Shooting RAW is the most efficient way to shoot." And then hit them with the "It's 4k, you know."

Graeme

Justin Anderson

10-04-2007, 07:10 AM

Joe Vinson

10-04-2007, 07:16 AM

Correct me if I'm wrong, but 4:4:4 is the ratio of data retention in YCrCb systems. RED uses RGB, not YCrCb, right? So on that basis, it's true: RED doesn't shoot 4:4:4.

But in the sense that 4:4:4 acquires a picture where no element has been sacrificed to another element for bandwidth reasons, RED fits the bill.

Is that accurate, or have I mucked it up further?

Graeme Nattress

10-04-2007, 07:37 AM

The recorded sampling is the same as the sensor sampling, and in that respect no subsampling has occured. However, with a bayer sensor, (and there are very valid engineering reasons for this approach) the sensor only has 2 green photosites, 1 red photosites and 1 blue photosites per group of 4 photosites. Now, this is not as bad as it sounds as it sits behind an optial low pass filter which reduces incomming resolution so that the image is mostly (and you'll need a test chart to see any - normal images won't) alias free. From then on it gets complex as we reconstruct what the RGB image would be from that RAW Bayer pattern data, and we reconstruct to the equivalent of 4:4:4.

Graeme

Chris Kenny

10-04-2007, 07:40 AM

Correct me if I'm wrong, but 4:4:4 is the ratio of data retention in YCrCb systems. RED uses RGB, not YCrCb, right? So on that basis, it's true: RED doesn't shoot 4:4:4.

A system that captures 4:4:4 captures color information (chroma) at the same resolution as brightness information (luma), both horizontally and vertically. A bayer-pattern sensor like the one in Red actually doesn't do this. But the exact resolution of both the chroma and luma information captured by a bayer sensor is not trivial to objectively quantify and is dependent on many variables, including the content of the scene and the specific algorithm used to reconstruct RGB data from the RAW data. Also, there will be differences in the resolution of individual color channels, which there's no way to note in the 4:x:x notation.

The upshot is, as Graeme says above, it makes much more sense to completely avoid that notation and tell people the camera shoots RAW. There are enough people out there who have shot RAW on digital SLRs that this description should be understood fairly widely.

Graeme Nattress

10-04-2007, 07:43 AM

And, chroma subsampling dictated by 4:2:2 say, is a process where RGB is converted to YCbCr and then Cb and Cr are resolution reduced. This process does not occur in RED at all. You get exactly the same resolution as the sensor recorded - it's just that the sensor records resolution "differently".

Graeme

Jeremy Hughes

10-04-2007, 07:56 AM

Tell them that unlike 4:4:4, you control the image. Or: You don't just get the best possible colour, you control it too.

Gavin Greenwalt

10-04-2007, 02:14 PM

I tell them that I think what the red does is better than 4:4:4 but that I have no idea how to explain this to them.

Well it's definitely not better than 4:4:4 that would be like photographing the color that the real world wants to be.

I suppose if you had a camera that lit and graded for you it could be considered 5:5:5. ;)

Stephen Williams

10-04-2007, 02:40 PM

Hi,

4:4:4 refers to SD video, the term is often used for HD however that is incorrect.

Stephen

vanguy

10-04-2007, 02:59 PM

Would I be out if I said that Red RAW records in "better than 4:4:4" quality? I know that 4:4:4 (or any color balanced, encoded) signal is going to be a diminishment of what comes off the sensor. So it makes sense to me that Red RAW records all of what comes off the sensor (making allowances for the compression) and therefore has more information than 4:4:4.

Stephen Williams

10-04-2007, 03:04 PM

Chris Kenny

10-04-2007, 03:15 PM

4:4:4 refers to SD video, the term is often used for HD however that is incorrect.

The notation was originally derived from the ratios of horizontal refresh frequencies in analog SD, but these days it's perfectly legitimate to use it with digital and HD (and super-HD) signals. At any rate, it's widely used that way by knowledgeable in the field, which should be enough to call it valid usage unless one believes it's inherently invalid for terms to change meaning over time.

Stephen Williams

10-04-2007, 03:27 PM

The notation was originally derived from the ratios of horizontal refresh frequencies in analog SD,

Hi Chris,

That was my point. No doubt people will call Red a 35mm film camera too.

Stephen

Chris Kenny

10-04-2007, 03:28 PM

Would I be out if I said that Red RAW records in "better than 4:4:4" quality?

It depends. In addition to using the notation to refer to the relative resolution of luma and chroma information, some people also use it to indicate oversampling. Under this scheme, a system which, for instance, captured both luma and chroma at 1280x960 pixels for an eventual 640x480 deliverable would be "8:8:8".

If you choose to use this notation to indicate oversampling, Red could conceivably be "better than 4:4:4", depending on what your deliverable resolution is. I'd recommend against trying to indicate oversampling like this, though, as there are much clearer ways to explain the same thing.

Graeme Nattress

10-04-2007, 03:39 PM

The problem is were trying to shoehorn an old piece of video terminology into a new area. It doesn't quite fit. It does not account for raw.

vanguy

10-04-2007, 04:07 PM

in my experience, when people (like producers) are asking for 4:4:4, they don't really know what it means, technically, precisely. They only know that it's the highest quality of video.

If they're little more technically inclined (like editors) they may be wanting to know how flexible it is. You can certainly push a 4:4:4 image around further than a 4:1:1 image as far as color correction goes.

One other thing they might be concerned with is how well it does with green or blue screen shots. There, again, 4:4:4 is best. Or so they've been taught.

So, if we can allay those fears, if Red RAW can do these things as well or better than 4:4:4, then clients will have a little more confidence. But if we talk about Not Really 4:4:4, or RAW SLRs or Bayer or subsampling, they're going to frown and maybe go away. They've been taught 4:4:4 is the best.

I know Red RAW is better for flexibility, and from what I've seen, you can't beat it for image quality. I don't know about greenscreen yet. I'd like to be able to say, "Red RAW is IT!" and that's why I'd like to tell them "Better than 4:4:4". I don't know if the majority of even industry professionals can appreciate the finer points.

dvpixl

10-04-2007, 04:22 PM

The problem is were trying to shoehorn an old piece of video terminology into a new area. It doesn't quite fit. It does not account for raw.

yes, perhaps it is time to rethink the process.

Shane Betts

10-04-2007, 04:31 PM

The easiest way to defeat the rental house argument is to just keep repeating at every dumb question, "It's not a video camera, it's a digital movie camera".

This is not a tarted up ENG camera and when they say it's 4k, it's actually 4k, not like those video cameras (try to place as much disdain on the V word as you can muster) that don't actually capture 1920x1080, more like 1440x540 (interlace/PsF).

In other words, take the high road. If they don't get it that's their problem. They'll understand one day.

Mark L. Pederson

10-04-2007, 07:09 PM

The problem is were trying to shoehorn an old piece of video terminology into a new area. It doesn't quite fit. It does not account for raw.

I like to say it's .... 6:6:6 ... the Number of the Beast

Emery Wells

10-04-2007, 07:22 PM

Shooting RAW is better than any 4:4:4 camera out there. There I said it, now you can tell all your clients that. :-)

Stephen Williams

10-05-2007, 12:06 AM

Shooting RAW is better than any 4:4:4 camera out there. There I said it, now you can tell all your clients that. :-)

Hi,

Until back to back testing is done, we don't actually know that for sure at this stage in the game.

Stephen

Jonathan L. Bowen

10-05-2007, 01:42 AM

Sure we do. We know there is no 4:4:4 camera out there that can come close to what the RED One can do, so that's good enough for me.

Stephen Williams

10-05-2007, 01:55 AM

Sure we do. We know there is no 4:4:4 camera out there that can come close to what the RED One can do, so that's good enough for me.

Hi Jonathan,

At the moment there are 4:4:4 cameras that can record upto 60 FPS. AFAIK RED can not do this today. Once all features are enabled & firmware issues are sorted Red well may have an advantage. Remember Red is a new camera, only 50 cameras delivered to date.

Stephen

Mark L. Pederson

10-05-2007, 04:25 AM

Hi Jonathan,

At the moment there are 4:4:4 cameras that can record upto 60 FPS. AFAIK RED can not do this today. Once all features are enabled & firmware issues are sorted Red well may have an advantage. Remember Red is a new camera, only 50 cameras delivered to date.

Stephen

our Red cameras record higher than 60 fps - right now.

Craig W. Bickerstaff

10-05-2007, 04:40 AM

Sure we do. We know there is no 4:4:4 camera out there that can come close to what the RED One can do, so that's good enough for me.

Really, and what is that exactly.

Rob Lohman

10-05-2007, 06:56 AM

I like to compare it to shooting RAW on a digital SLR. Then I say we don't drop any samples (like you would do in a 4:2:2 recording), which is 'like' 4:4:4.

The important bit of information is that we don't throw anything away before the compression. Obviously the compression throws some things away, but that's far more sophisticated than a simple dropping of samples (like 4:2:2 systems do).

Graeme Nattress

10-05-2007, 07:37 AM

We keep what the sensor records, and then use modern compression techniques on that. That's much more sensible than using old compression techniques on data that's half there.

Graeme

Jay A. Kelley

10-05-2007, 09:28 AM

In a way I am going with Stephen on this one. By RED being a "New Camera" I take that to mean there is not a lot of "aired footage" of it in the world right now. No one will give a rat's scrotum what the numbers are if the picture is BEAUTIFUL! And we are a lucky few who know just how fantastic the picture can indeed look.

And don't forget kiddies, for those who HAVE to have pure perfection, RED does have a RAW port out there somewhere.

Numbers are fun, but it's the look that matters.

Jay

(Rat's scrotum?! Where in the cornbread hell did THAT come from. Don't eat Ho-Hos before 11am)

David Battistella

10-05-2007, 11:23 AM

We keep what the sensor records, and then use modern compression techniques on that. That's much more sensible than using old compression techniques on data that's half there.

Graeme

To me, this is the whole point. A 4:4:4 signal is never going to get better than on the day it was recorded, but because you can record and compress the whole sensor's data and improve on that over time (with future REDCODE software releases) that it what puts RED ahead.

Essentially, the day could arrive when an algorithm that is even better than the day it was shot. The only thing that compares to this is scanning film at 2K 4K or 6K. Film is the only other technology out there where the digitized picture can be improved upon over time.

REDCODE RAW footage can be better by the time you online compared to the day on the set.

What digital solution offers you that flexibility? Everything else, "is what is is" forever.

David

mdo

10-05-2007, 12:20 PM

Subsampling schemes reflect the ratio of data loss for purposes of compressing a signal into a channel of limited bandwidth. 4:2:2 indicates that there is a reduction of 50% of the chroma components in relation to the luma component. This is accomplished by DISCARDING pixel information.

So when someone asks, "Is it 4:4:4," they're asking, "Does the camera retain all of its pixel information." I would think that it would be appropriate to answer with a "yes" -- and then follow up with, "But the concept of subsampling isn't really applicable to Red, since it's a quantum step beyond technologies that are typically described in those terms."

Graeme Nattress

10-05-2007, 12:29 PM

Isn't a quantum step the smallest possible step though :-)

Graeme

Chris Kenny

10-05-2007, 12:43 PM

So when someone asks, "Is it 4:4:4," they're asking, "Does the camera retain all of its pixel information." I would think that it would be appropriate to answer with a "yes" -- and then follow up with, "But the concept of subsampling isn't really applicable to Red, since it's a quantum step beyond technologies that are typically described in those terms."

Unless they're asking "Does the camera provide full resolution color data?", in which case the answer is no.

Just answer with "It shoots RAW, like a digital SLR. It outperforms 2K 4:4:4 cameras."

GlennChan

10-05-2007, 12:49 PM

Some different ways of looking at it:

- The quality you get from the camera is determined by the sensor and the compression/recording. So you have to look at both aspects... a sensor with very gentle compression can outperform an inferior sensor with no compression. Which looks like the case with Redcode RAW footage compared to some other uncompressed systems (e.g. modded DVX).

- For pixel-shifted 3CCD cameras, recording 4:4:4 R'G'B' (e.g. onto HDCAM SR) would actually be throwing some data away.
If you look at the modded DVX100, recording RAW lets you output a 1540x984 image that holds more information than a 4:4:4 720x480 image.

- A very good compressor would record all the information the sensor is outputting without discarding any useful information. The red RAW output port lets you record exactly what's coming off the sensor- you can't get any better than that. *Though of course it makes more sense to use Redcode RAW recording when the compression is so gentle (the difference is negligible) / visually lossless.
---Whereas if Red recorded 4:4:4, it would be throwing a lot of information away that you can't get back.

- From RAW you can derive a 4:4:4 image. And you have flexibility to do what you want with it in post.

- At medium-high bitrates, 4:2:2 is just bad/very inefficient compression.

GlennChan

10-05-2007, 12:58 PM

Unless they're asking "Does the camera provide full resolution color data?", in which case the answer is no.
Pixel-shifted 3-chip cameras don't provide full resolution color data either.

number6

10-05-2007, 01:51 PM

Don't eat Ho-Hos before 11am)

Ho had 'ho's? I thought he was monogamous. My esteem of the departed has just...risen.:clown2:

Stephen Williams

10-05-2007, 03:41 PM

our Red cameras record higher than 60 fps - right now.

At 4K ?

Stephen

mdo

10-05-2007, 09:15 PM

Good points. It's definitely not a 4K 4:4:4 camera. Interesting that a designation which used to be the Holy Grail has now become passé.

Maybe the correct response to Purefilm's questioners is, "4:4:4, are you kidding??? 4:4:4 cameras are OLD technology! The Red keeps a lot more data and is a lot more flexible than those things."

Gavin Greenwalt

10-06-2007, 12:57 PM

In my mind 4:4:4 = RGB. RED is not 4k RGB nor is it as good as 4k RGB. Therefore I say "Less than 4:4:4".

After all the future of RGB signals is such that in the future we'll be able to dramatically enhance their quality as well using OFlow Detail enhancement, so RAW isn't alone in riding the digital image processing wave.

Graeme Nattress

10-06-2007, 01:25 PM

But any camera system with 3 chips would still need an optical low pass filter ahead of those chips so that nasty aliasing does not occur, and that would too limit their resolution below 4:4:4, even though that's what gets recorded, which is somewhat innefficient.

I remain to be convinced onthe optical flow "super-resolution" for detail enhancement..... Super-resolution algorithms generally require aliased data :-)

Graeme

Jim Arthurs

10-06-2007, 02:42 PM

I've managed my expectations from day one, and am completely pleased with what's coming my way, partially because I defined my expectations this way;

... All I want at the end of the day is to deliver a product at 2K or 1080 resolution, where each and every pixel is unique in RGB (which is my artist's approach to re-defining 4:4:4). Oh, and I want it to be super smooth without aliasing and "naturally sharp" at this resolution. And cost less than any other solution out there, by a factor of 10 if possible. Thanks to RED, done deal.

The workflow to get what I want from RED is simple, it happens by shooting 4K and downsampling. Is this point actually reached at 3K, 2.5K, or 2K for the final downsample? I don't know and really don't much care. I suspect that there's quite a bit of wiggle room above 2K to satisfy me for cropping and repo work and still being tack sharp at 2K.

All I know is that this whole concept of downsizing Bayer data seems MUCH better than starting off with a camera that tops out at 2K to begin with, whether Bayer or 3 chip.

Graeme Nattress

10-06-2007, 03:49 PM

There is definate validity to the oversampling approach. Also, there's validity to properly filtering so that nasties don't get into the system even if you don't oversample. To me, I'm happy that even at a per-pixel level you get a smooth, continuous image that looks good even when zoomed pixel for pixel on your display.

Graeme

GlennChan

10-06-2007, 03:50 PM

In my mind 4:4:4 = RGB. RED is not 4k RGB nor is it as good as 4k RGB. Therefore I say "Less than 4:4:4".
This is getting into semantics, but IMO that definition of 4:4:4 is not that useful.

Suppose you had a 3-chip system, with each sensor having 4k photosites. Arguably, a 12K bayer-pattern sensor would yield better performance. Though there are tradeoffs between different sensor designs... 3-chip with no pixel shifting, 3-chip with pixel shifting, Foveon, and Bayer. e.g. depth of field, prism limitations, silicon being a poor color filter, etc.

2- Perhaps in practice it's not that useful to distinguish camera systems based on 4:4:4 versus RAW versus 4:2:2 (etc.).

Maybe something like this is somewhat better:
a- # of photosites and sensor design
e.g. _____ megapixel Bayer, 3 X 1080HD (or 3CCD 1920x1080)
And then you should know the relative tradeoffs between the systems... where 3-chip designs aren't necessarily that efficient compared to Bayer (though it might depend slightly on your application).

The downside of breaking down resolution-related performance this way is that it ignores a number of other factors. For Bayer systems, the quality of the de-mosaicing makes a big difference in any artifacts you get (e.g. look at stills from some of the other Bayer systems out there, especially the ones with less pixels/photosites).
And then there are minor differences in the amount of optical low pass filtering, what type of optical low pass filter is used, pixel shifted versus not pixel shifted, etc. Where pixel shifting makes tradeoffs between luminance and color performance.

b- Some measure of compression efficiency/quality
Granted, this is fairly difficult to put a number to that everyone would agree on. There's ways of fudging these figures.
e.g. 100mbps DVCPRO HD is supposedly many times better than 25mbps HDV.. but HDV compression is more efficient (ignoring the workflow implications of inter versus intraframe compression).

Omnius

10-07-2007, 02:35 PM

Your question cannot be answered.

RED does not shoot 4:4:4 or 4:2:2 which are video chroma sub-sampling terminology.

RED shoots RAW from a Bayer Patttern sensor. There's lots of talk on this forum about what that means.

Graeme

I think you should tell that to the people at Mundo Grúa (http://www.mundogrua.com/) who are representing the Red camera for rental as a 4:4:4 piece of hardware