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

[Adrian Jebef]

We would concede that Alexa has provided a nicer default image from its HD-SDI output than either RED ONE or EPIC. We've always looked at our HD-SDI output as more of a video tap than an end point because the .R3D file has the full gamut and precision of the sensor stored in it which the HD-SDI output just can't carry through rigorous post production.

However algorithms advance, so it would be a good idea to re-evaluate the HD-SDI default output of EPIC under Redgamma 3 / Redcolor 3 ...

Would it be correct to say that the Epic's 1080p HD-SDI output is not fully debayered?

 

[Eric Lange]

Yeah, maybe those engineers @ Sony don't have a clue what they are doing.

Yo dawg…

(that’s my Randy impression).

The illustrations you show are taken from a book called “Foundations of Vision by Brian A Wandell” ISBN 0-87893-853-2.

The illustrations you show are on page 49 on a chapter dealing with the “photoreceptor mosaic”.

The photographs that you show, shows the layout of Rods and cones, if you turn to page 53 (assuming you have the book... ) there is a two page discussion exactly as to why the S-cones (the blue sensitive ones are so Widely spaced) and how it is that there are basically NO blue receptors in the centre of the Fovea.

The photograph that you show of the centre of the macula shown, basically does not contain S-cones (blue sensitive receptors).

If you scrunch things together then a hexagonal arrangement is the most efficient cellular packing factor like that in a honey comb in a bee hive etc. Soap bubble foam etc.

The F65 arrangement is not a hexagonal packing factor but a rectilinear packing factor; the fact that it has been rotated 45 degrees does not make it hexagonal.

The reason for the diagonal arrangement is for what is called fast “zig-zag” read out times that would be more appropriate for near real time broadcast applications. There are many patents out there extolling the benefits of the zig zag read out arrangement. Has really very little to do with what the marketing would want you to swallow. I genuinely think this is a rather masterful attempt (by SONY) at making people accept a diagonally arranged Bayer array; (for patent reasons they had better not call what they have a rotated Bayer Array even if it is, otherwise their sensor is basically not novel/is obvious or cannot go to grant as a patent). Most punters instinctively would reject out of hand a diagonally arranged array of photo sites. So in that sense the marketing department has done a VERY good job. No one is complaining that diagonally arranged elements is super bad or ridiculous… which would be the normal “gut reaction” to the idea… so very cleverly they have shifted focus away from diagonally arranged elements as being “weird”.

I am not against the F65 at all, but I don’t like the smoke and mirrors and pull the wool over people’s eyes approach.

I say call a spade a spade… say WHAT IT IS… not what it is not…

I really want to see what the F65 can actually do.. rather than what the marketing guys are claiming it does.

Science and marketing historically has always been a bit of a disingenuous car crash.

So for me it’s not about SONY vs RED, to me it’s about Marketing guys versus Scientists and card carrying technologists. This problem crosses all boundaries and companies.

With great respect… to all,

Eric

 

[Peter Moretti]

...

Secondly, its blatantly obvious it was drawn that way to accentuate the benefits. That's the marketing thing.

...

I don't believe this is the case. The pictures compare a "Conventional Bayer Sensor" to the "New 8K Sensor of the F65." In fact, the F65 has a very high fill rate, which would mean decreased spaces between the pixels. And while the pixels themselves are square, they are covered by microlenses, which are round.

All that said, the F65 images I saw had some of the most wonk colors I've ever seen in my life. Of course, the camera is supposed to have been tweaked since then.

 

[Graeme Nattress]

Would it be correct to say that the Epic's 1080p HD-SDI output is not fully debayered?

No, that would not be correct.

Graeme

 

[Hrvoje Simic]

If you scrunch things together then a hexagonal arrangement is the most efficient cellular packing factor like that in a honey comb in a bee hive etc. Soap bubble foam etc.

Yes, the hexagonal arrangement was also the most logical packing factor for me as well, and I don't think we are the only ones working in this domain who had biology and geometry in elementary school.

I would propose exploration of both sensor and display technology in hexagonal arrangement, as it is the most natural, but it is logical to assume that this type of arrangement is pretty difficult to achieve.

The F65 arrangement is not a hexagonal packing factor but a rectilinear packing factor; the fact that it has been rotated 45 degrees does not make it hexagonal.

You are right, the arrangement is rectilinear and 45° rotation does not make it hexagonal, but the first published visualization of the pixel arrangement, in case it was not intentionally misleading, gave some clues in the principle in which the values are combined into one pixel.

Not sure about the pixel proportions, but if I recall correctly the principle was something like this:

If the values are combined in the way this visualization suggests, i.e. subdivided R&B pixels with cross diagonal combination, it gives a room to assume the principle is not just based on simple 45° rotation, but the different DeBayer logic as well, meaning not zig-zag.
Meaning symmetrical spacial offset of R&B to G.

 

[Gavin Greenwalt]

That would only be important if you did a 'dumb' debayer though that just bin'ed different photosites. I'm pretty sure every red photosite in RED's debayer takes into account every adjacent blue and green photo-site. I agree if you're just going to do pixel-binning without any fancy interpolation that takes into account adjacent pixels then the diagonal would be a superior means of binning. But that's not really 'debayering' so much as just grouping. The advantage of course is that you could chew through that with a GPU extremely easily.

 

[David Mullen ASC]

Yes, the hexagonal arrangement was also the most logical packing factor for me as well, and I don't think we are the only ones working in this domain who had biology and geometry in elementary school.

It seems logical though, that this type of arrangement is pretty difficult to achieve in sensor design.



You are right, the arrangement is rectilinear and 45° rotation does not make it hexagonal, but the first published visualization of the pixel arrangement, in case it was not intentionally misleading, gave some clues in the principle in which the values are combined into one pixel.

Not sure about the pixel proportions, but if I recall correctly the principle was something like this:

If the values are combined in the way this visualization suggests, i.e. subdivided R&B pixels with cross diagonal combination, it gives a room to assume the principle is not just based on simple 45° rotation, but the different DeBayer logic as well, meaning not zig-zag.
Meaning symmetrical spacial offset of R&B to G.

Wouldn't that design suggest that there were now twice as many blue and red photosites but that they were half the size of the green photosite?

 

[Adrian Jebef]

No, that would not be correct.

Graeme

G:

Uhmmmm... OK. So what would be correct? To my eye the Epic's HD-SDI output feels softer resolution wise than a full debayer out of DaVinci Resolve. Is the HD out a "video tap" or a full debayered and resized image?

 

[Graeme Nattress]

Adrian, probably difference in downsampling filters. The HD-SDI output is something we can continue to tune for greater performance though.

Graeme

 

[Adrian Jebef]

Wouldn't that design suggest that there were now twice as many blue and red photosites but that they were half the size of the green photosite?

I think it would. The design seems to attempt solving the problem of finer Red and Blue resolution within a Super 35 chip size. An interesting approach to oversampling a Bayer sensor in all RGB primaries for its intended target resolution (4K).

So, is it 4:4:4?

 

[Martin Stevens]

G:

Uhmmmm... OK. So what would be correct? To my eye the Epic's HD-SDI output feels softer resolution wise than a full debayer out of DaVinci Resolve. Is the HD out a "video tap" or a full debayered and resized image?

I think it is a SCALING issue that is the difference.

Perhaps.....the in camera SCALING is not as nice as the RedCine Software SCALING.

???

Also is the Scaling on the HDMI = to the Scaling on the HD-SDI?

 

[Eric Lange]

Yes, the hexagonal arrangement was also the most logical packing factor for me as well, and I don't think we are the only ones working in this domain who had biology and geometry in elementary school.

I would propose exploration of both sensor and display technology in hexagonal arrangement, as it is the most natural, but it is logical to assume that this type of arrangement is pretty difficult to achieve.



You are right, the arrangement is rectilinear and 45° rotation does not make it hexagonal, but the first published visualization of the pixel arrangement, in case it was not intentionally misleading, gave some clues in the principle in which the values are combined into one pixel.

Not sure about the pixel proportions, but if I recall correctly the principle was something like this:

If the values are combined in the way this visualization suggests, i.e. subdivided R&B pixels with cross diagonal combination, it gives a room to assume the principle is not just based on simple 45° rotation, but the different DeBayer logic as well, meaning not zig-zag.
Meaning symmetrical spacial offset of R&B to G.

Nice post…

Would be cute if they were physically built, arranged and wired that way…

But unfortunately they aren’t…

Here, check this out this is the original “Bayer” Patent…

Well illustrated especially for 1975… (read document then read it again holding at a 45 degree angle… only joking) (he,he).

http://www.google.com/patents?id=Q_o7AAAAEBAJ&printsec=abstract&zoom=4#v=onepage&q&f=false

I’m busy for the next few hours but will dig up some of those zig zag citations perhaps. For broadcast that seems pretty clever…

SONY engineers do know what they are doing but overall they have different design intent also for a particular “market” intent.

Cheers,

Eric

 

[Graeme Nattress]

I think it would. The design seems to attempt solving the problem of finer Red and Blue resolution within a Super 35 chip size. An interesting approach to oversampling a Bayer sensor in all RGB primaries for its intended target resolution (4K).

So, is it 4:4:4?

Well, it's not oversampling though, is it? It's under-sampled on red and blue and just sampled on green. We still need to put one on a chart and see it's aliasing and moire performance, and measure it's resolution and MTF. Smaller pixels make for higher sensor MTF, so you've got to be even more careful about aliasing and optical filtering. Because the output red and blue pixel locations don't coincide with sensor red and blue pixel locations we're also into the same territory that they are with the 1/3 pixel offset we can see on the F35 sensor. Obviously proper demosaicing and filtering can deal with this, so again, we have to test and measure to see what actually happens. Previously it's only been cheap low-end cameras that have used this Rot45 Bayer arrangement, and as we know what can be quite poor at low resolution can work very well at a high resolution.

It's a 17.7mp sensor (not the 20mp quoted - it's 4096x2160x2), so about 20% more pixels than Epic, but that's an area measurement - linear measurement it's 10% more so it's like the difference between 1080p and 2k - not much.

But again, as I've said above and other posters have commented - we've got to test and measure, especially for resolution and aliasing performance.

Graeme

 

[Hrvoje Simic]

Wouldn't that design suggest that there were now twice as many blue and red photosites but that they were half the size of the green photosite?

First published visualization of F65 sensor, showed a 45° rotation with additional vertical/horizontal divisions, affecting only R&B pixels, if I recall correctly. Not sure if that was visualized exactly the way it is actually done. At this point it is just guesswork.

This rotation approach was taken from ClearVid CMOS consumer technology Sony introduced in 2006. Back then they used more percent of green photosites.
Here is the interview mentioning this technology. Visualization of those sensors differs from the F65 sensor in the amount of green photosites and lack of additional divisions, but the principle seems similar:

http://www.camcorderinfo.com/content/The-ClearVID-CMOS-Sonys-Chip-of-Choice.htm

CCI: Why tilt them 45 degrees? What is that doing?

KB: For one, it allows us to use larger pixels over a standard horizontal pixel layout. If we rotate the position into a diamond [shape], it allows us to increase the size of individual pixels but keep the same spacing between the center points of each pixel, so you don’t lose any depth of field or degrade color.

 

[Les Dittert]

Aliasing problems go away of you are doing super debayer, that is debayer temporally.
So far all the talk is derived from still photo processing. Those papers are easy to find
We have motion, which does add significant sampling information that can be recovered.
I will present samples soon. I have RGBG planes to work with, avoiding the current Red methods.

Well, it's not oversampling though, is it? It's under-sampled on red and blue and just sampled on green. We still need to put one on a chart and see it's aliasing and moire performance, and measure it's resolution and MTF. Smaller pixels make for higher sensor MTF, so you've got to be even more careful about aliasing and optical filtering. Because the output red and blue pixel locations don't coincide with sensor red and blue pixel locations we're also into the same territory that they are with the 1/3 pixel offset we can see on the F35 sensor. Obviously proper demosaicing and filtering can deal with this, so again, we have to test and measure to see what actually happens. Previously it's only been cheap low-end cameras that have used this Rot45 Bayer arrangement, and as we know what can be quite poor at low resolution can work very well at a high resolution.

It's a 17.7mp sensor (not the 20mp quoted - it's 4096x2160x2), so about 20% more pixels than Epic, but that's an area measurement - linear measurement it's 10% more so it's like the difference between 1080p and 2k - not much.

But again, as I've said above and other posters have commented - we've got to test and measure, especially for resolution and aliasing performance.

Graeme

 

[Hrvoje Simic]

This visualization shows two options with differences in pixel size, simulating pixels arrangement with micro lenses.
Space between pixels on higher resolution examples is unintentionally smaller, as it was faster to just scale the first reference visualization I made, so this is merely a rough comparison.

First row from the left - resolution starting point, next simple double density, next same double density rotated 45°
Second row - starting point, next 45° rotation & proportions as Sony previously visualized, next pixel size comparison of that approach

In case the visualization previously published by Sony does not suggest additional pixel sub-divisions at all,
the point was to allow the analysis of this approach in general, regarding chroma values per same surface, and demosaic options.

At least of the principle...before the actual pixel-peeping.
For those who care that much with the level of picture quality which Red cameras achieve today.

 

[Peter Dmitriyev]

how does f65 have so many pixels/photosites but still able to get supposedly 13.5 or 14 stops dynamic range? I thought the smaller the photosites the less dynamic range so for instance red's 5 or so micron against alexa 7 and change micron photosite sizes makes sense for the 2 stop dynamic range difference yet much lower resolution in alexa...how does f65 go around this getting best of both worlds, higher DR than red AND higher resolution?

 

[Tom.Wong]

how does f65 have so many pixels/photosites but still able to get supposedly 13.5 or 14 stops dynamic range? I thought the smaller the photosites the less dynamic range so for instance red's 5 or so micron against alexa 7 and change micron photosite sizes makes sense for the 2 stop dynamic range difference yet much lower resolution in alexa...how does f65 go around this getting best of both worlds, higher DR than red AND higher resolution?

A. downsampling increases DR and reduces noise. their quoted "double the resolution" helps a lot with that going down to a 4k.

B. the material where the photosites are and light hits is usually made of aluminum i believe in most cases. so the transfer of energy is at a certain distance. if i remember correctly they use copper in their sensor so it's 50% less distance for the conversion. allowing the light to come in faster, thus increasing DR and lowering noise yet again, along with increased sensitivity.

C. it uses a quad based A/D conversion. the chip has 4 quadrants with 4 separate processing units outputting the data of just that quadrant, and goes through double noise reduction before it hits the card.



and with the combination of all these things, with these being some of the major factors. they have a claimed 14 stops of DR, a very clean 800 asa setting with a very low noise floor at higher iso's.


but remember the camera's price, and size too. pro's and cons to every method and cameras.

 

[Bob Gundu]

Ok guys... let's let Graeme get back getting the new gamma's out the door ;-)

 

[Peter Dmitriyev]

A. downsampling increases DR and reduces noise. their quoted "double the resolution" helps a lot with that going down to a 4k.

B. the material where the photosites are and light hits is usually made of aluminum i believe in most cases. so the transfer of energy is at a certain distance. if i remember correctly they use copper in their sensor so it's 50% less distance for the conversion. allowing the light to come in faster, thus increasing DR and lowering noise yet again, along with increased sensitivity.

C. it uses a quad based A/D conversion. the chip has 4 quadrants with 4 separate processing units outputting the data of just that quadrant, and goes through double noise reduction before it hits the card.



and with the combination of all these things, with these being some of the major factors. they have a claimed 14 stops of DR, a very clean 800 asa setting with a very low noise floor at higher iso's.


but remember the camera's price, and size too. pro's and cons to every method and cameras.

yeah who knows I never even seen it tested yet, their 14 stop claim might be all bunk as far as we know.


but just out of curiosity are you saying if red MX sensor was an 8k sensor and down sampled to the 4.5k then maybe it would have higher DR etc?