Is R3D 4:4:4 or 4:2:2?

[Adrian Jebef]

I see no discussion, that's what my comment refers to.

Where this thread goes next is up to you. If it's a statement that we shall just have to agree to disagree, we can respect that.

So you won't even share a beer with me? Ah, man, come on. It's a Friday!

 

[Adrian Jebef]

So what does that mean to us layman who are only viewing the final produced image with our eyes and not intellectually as a series of numbers? Does this mean according to one of the last posts that the Sony F35 which does true 4:4:4 has basically greater and truer color fidelity than Red MX/Epic etc? Does this mean it's colors are truer, more accurate, richer, what? Does this explain Red's micro-tone separation problems on skin tones etC? Just wondering.

I'd say that the way color resolution is handled in a R3D files does present some problems like the one you mentioned concerning skin tones.

 

[Adrian Jebef]

I hardly think that the F35's particular color filter array makes it significantly more "true" 4:4:4 than something with a bayer pattern, it is still an array and any one photosite still only has one color filter on it. From what I understand, the C300 does a similar thing with its Bayer CFA in that it treats quad pixels as RGB pixels, R/G/B are still not co-sited. I would tend to think that these cameras that skip the demosaic/debayer step aren't giving you more color information, they just give you less luminance information since they do not attempt to reconstruct or interpolate missing RGB data from the full pixel array. So maybe the F35 is 4:4:4 in 1080p terms, but using that same terminology in 4k terms, the F35 delivers 4k 2:2:0, whereas RED delivers 4k 4:2:0 (sorry Graeme for perpetuating the misuse of 4:4:4 etc outside of its intended YUV sampling realm, but for simplicity rather than accuracy it seems to be something people have an easier time understanding for comparison). Perhaps all you'd have to do to get back F35-like 4:4:4 from RED 4kHD is do a nearest-neighbor downscale to 1080p? Or maybe RED will eventually release their 1080p RGB recording mode mentioned years back, and maybe it will just read quad pixels as RGB pixels without debayering, but why would we want that over 4k debayered and downscaled?

Noah:

As far as I am aware you cannot skip the demosiac / debayering process. Otherwise you would not have an RGB image. There definitely are different ways in which to debayer the same RAW sensor data to varying degrees of success. The early days of the RED One and changes in RED's R3D processing are notably the best example.

To be technically correct for our Workflow Wizards and Problem Solvers, a Bayer color filter array (CFA) is made up of 50% Green photosensors, 25% Red photosensors, and 25% Blue photosensors. This is known as RGBG and can perhaps be expressed as 4:2:2:4 So should we agree that the Epic's R3D file is 4:2:2:4? That would be cool. I wanna get that printed on a t-shirt right now!

 

[Peter Dmitriyev]

A shame Cus id gladly take 2k with better dynamic range and color fidelity over 4k+ etc but then I guess id just have the alexa and there's a reason I suppose that it costs 3 times more than an mx or epic

 

[Evin Grant]

A shame Cus id gladly take 2k with better dynamic range and color fidelity over 4k+ etc but then I guess id just have the alexa and there's a reason I suppose that it costs 3 times more than an mx or epic

Absolutely not, the Alexa has a nice output image but in terms of pure color rendition the Epic wipes the floor with it. Including skin tones. However with more information comes greater responsibility and the need for greater understanding of the material and processes required to control and manipulate said information. If you like your images spoon fed to you the Alexa/C300 etc. are a good fit. If you want full access to what is really there the only choice is Epic.

My second contention is that saying a GRBG Bayer filter array is 4:2:2 or 100%/50%/50% is like saying that a 4 cylinder is 30% less powerful than a 6 cylinder. Despite the fact that the 4 cylinder is a tuned Subie STI and the 6 cylinder is a Camry. Debayering is a computational act, it uses real, interpretive and calculated information to create color channels that measure much higher than the 4:2:2 or 100/50/50 you imply because no color is absolute, it is made up of components and those can be calculated to a much higher degree of accuracy. Measured chroma resolution here is the important thing to consider because that is the chroma resolution actually available to you when you debayer and process your footage. Is in absolute terms the color resolution lower than the luma? Yes, but by only a small margin and at least at 5K you'd be getting around 4K or more, again tested, usable, measured resolution, not theoretical numbers based on a sub-sampling method that doesn't even apply to RGB. Remember YUV is in and of it's self a chroma subsampling method.
Simply by staying in the RGB space you've saved significant color information from being thrown down the drain.

Put into YUV terms if the Alexa produces a pure 4:4:4 image (It doesn't but I'll go with it) then the Epic produces a 20:16:16 image, which would you choose?

 

[Graeme Nattress]

I'd say that the way color resolution is handled in a R3D files does present some problems like the one you mentioned concerning skin tones.

But you're saying Alexa is better, and as we've just discussed that camera has a lower colour resolution.

Peter: As for the F35, it may appear that it does "True 4:4:4", but when you actually look at the image it's corrupted with aliases and chroma moire. I don't see how such artifacts make an image in any way better. It's pretty obvious from any diagram of the F35 sensor that when you make an image that the red and blue pixels are 1/3 of a pixel horizontally mis-aligned from the green. This is what leads to the chroma rainbows. The chroma rainbows are sampling error colours that don't exist in the target.

Graeme

 

[Wayne Morellini]

Graeme:

This thread was not started to discuss the use of proper terminology. The truth is that our industry's terminology is not correct or incorrect but simply in a constant state of flux. 4:4:4 was coined when we may have both been in grade school. Technically, the term implys a specific technical conversation. However, in 2012 the actual importance of full RGB color sampling (4:4:4) has changed relative to emerging thechnology like RED's R3D file format. So the question still begs to be answered: How does the machine work? Does the R3D contain full luma resolution and full RGB color resolution (fidelity)?

Wayne is on target here when he says "Bayer is adequate if you treat it right, especially downsized right, but not as good as triple sampling. My pixel shift handled right might be suitable, but less accurate again, at 8k+ it may not be much of an issue." Summed up any Bayer-type CMOS sensor can get close to proper 4:4:4 full color sampling by over sampling luma and chroma resolution with the intention of downsizing to an intended target resolution. Sample at 4K and conform to 2K/HD. The ARRI Alexa, Sony F65, and Canon C300 work this way. The Epic is designed to conform for max luminance resolution regardless of color. The debayering and software conversation process can be tweaked to make color reproduction seem more complete but it can never add discreet RGB information that was never sampled at the sensor to begin with.

Man, guys, come on. I wasn't going to post again, but it is often bewildering these sort of duscussions, because they often feature obvouse misunderstandings.

Of course 4:4:4 has the "ability" to present more accurate color fidelity when handled right, but this is different from color spatrial resolution (bits per pixel of accuracy versus pixel resolution of color).

The 4:4:4 thing over Red Bayer does not matter so much, because of Red's industry leading debayering, and the new color science geals many things (thanks Graeme) though I have not seen any actual facial example shots over a variety of lighting conditions and skin tones. If you look into the underworld 4 thread mainly I conjecture about the too obviose skin tone problems, I forget what I wrote now, but have a look there. This problem seems to be getting better and better, and I hazard a guess is now dealt with, You get a nice ussable colored image of suitable quality, so the 4:4:4 to Bayer issue is more of an non issue unless you want very high grade (but then you have the problem if how to get a 4:4:4 resolution 4k camera or sensor of high quality, like Graeme said, which I have solved and a camera company has a sensor design that matches my previouse design, which would be a matter of history, except Chris's dvinfo blocked my post on it. So yes, you should be able to maximise 4:4:4 in everyway with my designs Graeme. But the reality guys, is my design is not available to motion yet, I think, and Bayer is and is probably superior to what 4:4:4 solutions are available, therefore Bayer wins (for now).

Red is trying to deliver the best picture to work with, less making it too much like film. It is better to have 5k of quality pixels then 2k 4:4:4 of a mess, simply. So, it is better to have quality true 5k bayer than 1080p 4:4:4 output from the same camera as well, to play with in post. 4:2:0, 4:4:4 is goid fir live briadcadt, where there is no post, Viewing is a problem though, unless they can get on set live debayering and color correction to output 4:4:4/4:2:2 to view. This maybe unlikely to happen soon, as the processing step might add too much of a delay in transmission. However, my mind is devising a single function parrallel debayer circuite idea here on the fly. The actual circuites could be simple, especially if we can do analogue approximations of the digital functions we want to replicate it in post. Delays could be taken down to a cycle for however many pixels the curcuite is designed to handle (but this is wishfull thinking in dealing with analogue). So in the meantime, before optical computing or this analogy idea, you might have to do with a rough debayer or delayed quality one, and learn to read Bayer pixels on another monitor. : (

Bayer can't really even be more accurately described as 2:1:1.

The terms used are technical hard facts, that are nearly unchanging unless they are further developed. 4:4:4 is the same in it's varieties. Technical people are the people to judge accuracy, not what somebody thinks they see or likes, they are different aspects. People can think something is more sharp/accurate due to sharpening for one, or just contrast preference (all controversal I know) just as examples. But better to start with the better image before you add these. We then go into all the other technical aspects Graeme talked about.

The film industry is very small, and eventually most everybody that wants and can afford a Red may get one, then they will have mostly nothing left to buy until it dies or they upgrade over the next ten, maybe twenty years. Red has to watch it's step to avoid a sales slump. From what I can tell, the broadcast, videographer and enthusuast (prosumer), photigraphy market is maybe neglected, this is where you can get the majority of sales, and ongoing replacement on death sales and resales once the cinema dustry is full up, to keep production numbers up to keep price down. Imagine trying to survive on 100-1000 cinema camera replacements or resales a year, versus 10,000. I have been watching this unfolding for years now. Not a problem, hopefully, if you home cameras handling and features for this markets as well.

There maybe others I have forgotten, still sick. I understand where Graeme is coming from, we look and think sort of alike, except I'm like Graeme on steriods. The well honed ideas going on in my head makes Red look like a girls kindly garden. I know some of of my statements might be a bit provocative sometimes, and Graeme is such a lovable character, but I am trying to get Red to take notice, think and come over and see what's happening. What can really be done beyond their "imagination" as Jim puts it. I have invested confidence in Red that they can do the job of lower cost cinema cameras, and with quality. I have been waiting for some issues to be solved, but that appears to be happening, just lower end models and cross market features to go (well apart from other announced products). Though, I have noticed Graeme avoiding addressing the ideas in my posts. Remember, not everything is as it seems, what has been attempted before is not evidence of what can be done. What might not have worked before does not mean it cant be engineered around in future. It is a shame I have forgotten so much.

A quote: When somebody thinks you are a fool, they often do not follow what you are saying and miss what is going on, and become the fool, because even a fool may speak wisdom, and even the wise appear a fool (and the listerner miss wisdom,) if one does not listen for wisdom. It is better to look for the wisdom than solely the foolishness, so we might learn something. (from each other)
Me. brackets alternative versions.

 

[Wayne Morellini]

Graeme, could we please see your diagram of the f35? I don't remember it, and don't know if I have the right pattern inside my head or another.

Thanks.

 

[Graeme Nattress]

Bad article, but the image of the F35 sensor is correct: http://notesonvideo.blogspot.com/2011/01/sony-f35-another-unusual-sensor.html

And this is the lovely chroma moire: http://nattress.com/F35chroma.png

Graeme

 

[Wayne Morellini]

Thanks Graeme, this is not the sensor I was thinking about, that had a diagonal tilted array like their, low end cameras have gone over in recent years, which where a bit harder to appreciate (though what I said still stands, interesting twist) but this makes it easier. I thought the video industry went over to Bayer because it was better than rgb stripe {: ( . However, from what I remember I did not notice anything in either Transformers 3 or Tron, apart from some low color action camera footage from "freinds". So maybe they are depending on demosaicing it out, moire and ant-alaising software.

I have just been looking around on some of the subjects here and found some nice examples before. They illustrate something I was pointing out about the fovein and 100% fill factor. Obviously the microlens in the newer foveon are not going to be 100% coverage perfect and still cause some issue, but near 100% should now be possible by other means. You should be able to use software for moire removal, as debayering can do for anti-alaising, and matter of fact the foveon has advantages over Bayer in this respect. The Bayer situation in general, excluding your software, is so bad a low res foveon gets favourable comparisons. I should not have discounted them so quickly, but they don't do HD+ video well.

www.en.wikipedia.org/wiki/Foveon_X3_sensor:
There is a little more in the original.

However, the actual resolution produced by the Bayer sensor is more complicated than the count of its photosites, or its native file size, might suggest. The reason has to do with both the demosaicing and the separate anti-aliasing filter commonly used to reduce the occurrence or severity of color moiré patterns that the mosaic characteristic of the Bayer sensor produces. The effect of this filter is to blur the image output of the sensor, thus producing a lower resolution than the photosite count would seem to imply. This filter is largely unnecessary with the Foveon X3 sensor and is not used. The earliest camera with a Foveon X3 sensor, the Sigma SD9, showed visible luminance moiré patterns, but not color moiré.[25] Subsequent X3-equipped cameras have less aliasing because they include microlenses, which provide an effective anti-aliasing filter by averaging the optical signal over an area commensurate with the sample density, which is not possible in any color channel of a Bayer-type sensor. Aliasing from the Foveon X3 sensor is "far less bothersome because it's monochrome" according to Norman Koren.[26] Therefore, in theory, it is possible for a Foveon X3 sensor with the same number of photodiodes as a Bayer sensor and no separate anti-aliasing filter to attain a higher spatial resolution than that Bayer sensor. Independent tests indicate that the "10.2 MP" array of the Foveon X3 sensor (in the Sigma SD10) has a resolution similar to a 5 MP[27] or 6 MP[28] Bayer sensor, and at low ISO speed even similar to a 7.2 MP[29] Bayer sensor.

With the introduction of the Sigma SD14, the 14 MP (4.7 MP red + 4.7 MP green + 4.7 MP blue) Foveon X3 sensor resolution is being compared favorably by reviewers to that of 10 MP Bayer sensors. For example, Mike Chaney of ddisoftware says, "the SD14 produces better photos than a typical 10 MP dSLR because it is able to carry sharp detail all the way to the 'falloff' point at 1700 LPI whereas contrast, color detail, and sharpness begin to degrade long before the 1700 LPI limit on a Bayer based 10 MP dSLR."[30] Another article judges the Foveon X3 sensor as roughly equivalent to a 9 MP Bayer sensor.[31] A visual comparison between a 14 MP Foveon sensor and a 12.3 MP Bayer sensor shows Foveon has crisper details.[32]

www.en.wikipedia.org/wiki/Anti-aliasing_filter

The choice of spot separation for such a filter involves a tradeoff among sharpness, aliasing, and fill factor. In a monochrome or three-CCD or Foveon X3 camera, the fill factor alone, if near 100% effective with microlenses, can provide a significant anti-aliasing effect,[2] while in color filter array (CFA, e.g. Bayer filter) cameras, an additional filter is generally needed to reduce aliasing to an acceptable level.[3][4][5]

It is interesting, that years before I knew about how these filterung systems work, I devised a plan to get rid of jaggies, flashing pixels, increased comoressiin, and debayering color space accuracy for a camera that had these issues, just defocus the image just enough so the light from pixels would encrouch on neighboring pixels. Like a filter. But the interesting thing, is that the spread of light of a pixel allowed it hidden primaries to be shared and registered with it's neighbors.

The problem with resolution tests are, that even though the light. is so spread out by the filters etc, blurring, the pixels are still registering some of the original pixel data, so is resolution ever entirely lower than the pixel resolution?

 

[David Mullen ASC]

These theoretical discussions are interesting... but let's get real here -- there is no "perfect" imaging system (and human vision isn't perfect either). Comparing Red or Alexa, etc. to any theoretically "perfect" system or device has very little practical use other than perhaps a starting point for building future "less imperfect" devices. And certainly film negative isn't perfect either -- even if color information is captured by various layers, that doesn't mean each layer has equal resolution, just as when you scan a film image at 12K, that doesn't mean it contains 12K worth of detail.

I suppose in theory a 3-sensor camera would capture color information more "equally" per channel, but then you run into the weaknesses of relying on a prism-block design, plus the weight issue if you made each sensor 35mm in size. You could shoot an image three times in a row with different filters and use a single sensor, but of course that isn't very practical for motion picture work, let alone still images of moving objects.

All real-world single-sensor solutions end up with their own strength and weaknesses, both in image quality and the practicality and affordability of the camera... right now, most of these digital cameras are trying to maximize quality from a CMOS sensor with a Bayer CFA, perhaps the future will involve radically different sensor designs, and of course the color science behind all of this will continue to improve too, but I don't see the point in calling any RAW file "4:2:2" or anything like that, it seems misleading and confusing.

 

[Wayne Morellini]

We got real David, eventually (I assume all of us will be).

 

[Eric Lange]

But you're saying Alexa is better, and as we've just discussed that camera has a lower colour resolution.

Peter: As for the F35, it may appear that it does "True 4:4:4", but when you actually look at the image it's corrupted with aliases and chroma moire. I don't see how such artifacts make an image in any way better. It's pretty obvious from any diagram of the F35 sensor that when you make an image that the red and blue pixels are 1/3 of a pixel horizontally mis-aligned from the green. This is what leads to the chroma rainbows. The chroma rainbows are sampling error colours that don't exist in the target.

Graeme

Graeme you are just not getting it… those color sampling errors that don’t exist in the target are AUTHENTIC color sampling errors that don’t exist in the target… !!! (he, he)..

Goodness me..

Tut, tut,

Cheers,

Eric

The higher spatial res sensors Dragon+ will I think will embody a very key tipping point where such discussions in the future become moot and the conceptual notion of 4:4:4 becomes almost a “Dinosaur” or at least an intermediate “working theory” much like Phlogiston in the late 1600s. Six years from now no one is going to be having trouble with this and we quibble and agonize about some other new technical dilemma either from a comfy armchair, on set, hanging out of a helicopter or from the deepest depths of the ocean.

http://en.wikipedia.org/wiki/Phlogiston_theory



Let’s hope that we don’t need to dephlogisticate our RAW files in the future…

 

[Graeme Nattress]

Wayne, the Foveon white papers say that an OLPF is still necessary to reduce aliasing on their sensor. "This filter is largely unnecessary with the Foveon X3 sensor and is not used. " is opinion and not fact, and not backed up by the Foveon white papers themselves. Sure, you don't get chroma moire, but that's the easy bit to remove on a Bayer CFA. Luma moire is practically impossible to remove on any sensor type, especially on moving images.

"the fill factor alone, if near 100% effective with microlenses, can provide a significant anti-aliasing effect," - again, incorrect. A high fill factor will give the sensor a lower inherent MTF, but that's not the same as anti-aliasing.

Graeme

 

[Graeme Nattress]

Graeme you are just not getting it… those color sampling errors that don’t exist in the target are AUTHENTIC color sampling errors that don’t exist in the target… !!! (he, he)..

Yes, I'm sorry that I'm not getting it. Perhaps we should start engineering the cameras with obvious aliasing and chroma moire artifacts and use the "name badge reality distortion field" method of artifact removal.

Graeme

 

[Graeme Nattress]

but I don't see the point in calling any RAW file "4:2:2" or anything like that, it seems misleading and confusing.

Absolutely correct David, and well put.

"even if color information is captured by various layers, that doesn't mean each layer has equal resolution" - I remember the first time I looked at a film scan DPX and examined the individual channels and saw how different they can be. But it doesn't mean that film can't make a great picture, because it does! Same with RED Cameras. Although they use the often mocked (and mis-understood) Bayer Pattern CFA, they produce superb images in the right hands, and even I can make some pretty images with them too!

Graeme

 

[Eric Lange]

Yes, I'm sorry that I'm not getting it. Perhaps we should start engineering the cameras with obvious aliasing and chroma moire artifacts and use the "name badge reality distortion field" method of artifact removal.

Graeme

(he, he)…

That wouldn’t also be (by pure chance) similar to the “tilt your head 45 degrees to see the reality of the “Name Badge Reality Distortion Field” method of artifact removal reality… by any chance… also?

No reply needed…

Eric

 

[Graeme Nattress]

No reply given.

Graeme

 

[Uli Plank]

That wouldn’t also be (by pure chance) similar to the “tilt your head 45 degrees to see the reality of the “Name Badge Reality Distortion Field” method of artifact removal reality… by any chance… also?

F65 ?

 

[Eric Lange]

shhhhhh!

It seems Old ‘uncle Bayer” has a bit of a limp and is only alowed to “stagger” around in circles tilted on a 45 degree angle it seems…

Who will save him?