Color Reproduction and Scarlet

[digitalnirvana9]

I really like the HVX 200 because of the 422 color space. The colors are much deeper and nuanced than HDV and the composites are much cleaner IMHO. So my question is: I know that RED works outside of the usual color space system. Will Scarlet be able to reproduce the rich colors of the HVX if it uses a lesser rez system (2K?) but still bases its other attributes on RED for capture, editing and color correction? This might be a good question for Barry Green....Thanks, Norman

 

[Peter Majtan]

For Scarlet to be "truly" revolutionary I would expect it to capture 4:4:4 RGB, nothing less... I think, let me correct myself- I KNOW that the technology has matured enough to abolish "compromising" on color information...

Besides - native 4:4:4 RGB is much easier to process and actually needs less "electronics magic" to work. This goes even more for post...

The question we should be asking is if Scarlet (and a new version or RED ONE) is going to capture HDRI, or at least 16-bit (8:8:8)? This is not far-fetched, believe me...

 

[Stephen Gentle]

For Scarlet to be "truly" revolutionary I would expect it to capture 4:4:4 RGB, nothing less... I think, let me correct myself- I KNOW that the technology has matured enough to abolish "compromising" on color information...

Besides - native 4:4:4 RGB is much easier to process and actually needs less "electronics magic" to work. This goes even more for post...

The question we should be asking is if Scarlet (and a new version or RED ONE) is going to capture HDRI, or at least 16-bit (8:8:8)? This is not far-fetched, believe me...

There is no 4:4:4 - there is only RAW

 

[Bruce Allen]

For Scarlet to be "truly" revolutionary I would expect it to capture 4:4:4 RGB, nothing less... I think, let me correct myself- I KNOW that the technology has matured enough to abolish "compromising" on color information...

Besides - native 4:4:4 RGB is much easier to process and actually needs less "electronics magic" to work. This goes even more for post...

The question we should be asking is if Scarlet (and a new version or RED ONE) is going to capture HDRI, or at least 16-bit (8:8:8)? This is not far-fetched, believe me...

Peter

Two things: you can't get 4:4:4 from a single CCD, unless you are oversampling heavily (eg a Red One probably produces 4:4:4 2K).

4:2:2, 3:1:1 etc ratios do not mean bits per channel. They are just ratios of luma : chroma.

Agree, HDRI is hopefully in our future! Although it is currently too much of a tradeoff. I'd be happy with "just" being competitive with film.

Bruce Allen
www.boacinema.com

 

[Peter Majtan]

Peter

Two things: you can't get 4:4:4 from a single CCD, unless you are oversampling heavily (eg a Red One probably produces 4:4:4 2K).

Dude - RED ONE is not CCD - it is CMOS!!! (However this also makes no difference to 4:4:4)
Also I don't get Your point - You say You can't get 4:4:4 and You say "Red One probably produces 4:4:4 2K"

4:2:2, 3:1:1 etc ratios do not mean bits per channel. They are just ratios of luma : chroma.

Yes when You are talking about YUV (more precisely Y'CbCr). In RGB world 4:4:4 refers to 8-bit RGB and 8:8:8 refers to 16-bit RGB - just ask anyone with Sony Vialta film scanner. I was part of the team who pull the 8:8:8 data out of it...

Agree, HDRI is hopefully in our future! Although it is currently too much of a tradeoff. I'd be happy with "just" being competitive with film.

HDRI is not a tradeoff at all. I have designed and outlined HDRI capturing system capable of recording HDRI just like a "normal" camera's do. In-fact it is much easier the most people think. Just "the market is not ready yet" BS the big 5 is giving us is stopping HDRI from happening...

Jim - if You are listening - I will be more then happy to share my HDRI capture design with You guys, if You want to implement it to RED ONE, Scarlet, or any other shade of red You will develop in the future...



Bruce Allen
www.boacinema.com[/QUOTE]

 

[Peter Majtan]

There is no 4:4:4 - there is only RAW

Once again - 4:4:4 and RAW are two different things, altho clearly related. RAW is the format the "raw" data from the sensor are stored. 4:4:4 is the sampling of the data. Depending on the "RAW" format one could then extract different sampling data from it.

What really matters is how RED "interprets" and "stores" raw data from the sensor. Contrary to what others may think You don't need 3 sensors (altho it does helps heaps to have individual sensors capturing separate color channels) to get 4:4:4 out of a sensor - be it CCD or CMOS. Just look at still cameras - You get pure "4:4:4" or even "6:6:6" and "8:8:8" RGB still images - no subsampling what so ever! But that is another topic...

Given that Bayer pattern has double of green pixels as compared with red and blue - people tend to think that this somehow interprets as 4:2:2. It does and does not. First of all - in YUV world - 4:2:2 does not translate as 4R:2G:2B - which coincidently is the ratio of the colors in the Bayer pixel pattern. Here are the common systems in use today:

4:4:4 Y'CbCr
Each of the three Y'CbCr components have the same sample rate.
(no subsampling) - can be easily translated to RGB format without any loss of quality.

4:4:4 RGB implicitly does not have any chroma subsampling at all.
Some SONY formats can record 4:4:4 RGB over dual-link HD-SDI.
SONY Vialta can actually do 8:8:8 RGB (16-bit)...

4:2:2
The two chroma components are sampled at half the sample rate of luma, so horizontal chroma resolution is cut in half. This reduces the bandwidth of a video signal by one-third with little to no visual difference.

It is in the "Y'CbCr" world, where the decision was made to trade of chroma in favor of luma (due to data bandwidth limitations and human eye greater sensitivity to contrast rather the color). When capturing the luma (or brightness of the scene) - good quality cameras use data from all "four" (GGBR - having double of green pixels) adjacent sensor pixels, as color is irrelevant for the interpretation of the exposure to the sensor area for luma calculations. Also our eyes are more sensitive to green light, than to red or blue - this was the reason for Bayer pattern to be GGBR, not RRBG or BBRG...

I hope this clear things up a little...

 

[Peter Majtan]

Agree, HDRI is hopefully in our future! Although it is currently too much of a tradeoff. I'd be happy with "just" being competitive with film.

Just one more thing:

It is precisely HDRI (and only HDRI) that can make digital competitive with analog film. Digital has already matched (and exceed in case of 35mm) the resolution of an image. You can also use the "same" glass and any other accessories. The only remaining mountain to scale is the high dynamic range of analog film. Without HDRI we will never catch up...

 

[Graeme Nattress]

First, I'd argue against 4:2:2 giving better colour (deeper and nuanced) than HDV (which is 4:2:0) because contrary to popular usage of the term, 4:2:2 is NOT a colour space. A colour space is something like Adobe RGB 1998, or REC709, or sRGB etc. 4:2:2 is chroma sub-sampling notation.

I'm not going to talk about Scarlet here for obvious reasons....

First, 4:4:4 is very rare. Some cameras that claim it have offset sensors (pixel shift) which although can improve things at 4:2:2, makes things worse at 4:4:4. I've seen it in dpx files from said cameras where the colours just never seem to line up right, always having a pixel offset. Even cameras that do have 4:4:4 will find it compromised by the prism, lens etc.

The RED One has one of the nicest, most natural, unforced colour reproduction I've seen. I've obviously seen a lot of footage on my system over the years, from high end and low end cameras, and RED really does look very nice, and I'm not just saying that because of my involvement in the way RED ONE generates colour. The basic decode of colour on the RED ONE is very under saturated. That's because it captures a very wide gamut of colours, one of the benefits of our RAW approach. This is a good thing!

Chroma sub-sampling notation is a measure of by how much you reduce the resolution of the chroma components when the image is in YCbCr form. "least 16-bit (8:8:8)? ", chroma sub-sampling notation does not refer, in any way, to bit depth, so that question is meaningless. I see your comment that a scanner uses such notation through for 16bit transfer, but quite frankly, that's a pure abuse of the notation and not valid in any way. Remember when HD was referred to as 22:11:11 because of the higher sampling base frequency than SD?

Also, it is not correct or proper to attribute a RAW system with a chroma sub-sampling notation. That's because the Bayer pattern used functions completely differently to reduce the image data, and, as the light hits the sensor, rather than dealing with decimating transformed data after capture.

When we reconstruct RAW, we aim to reconstruct back to full RGB, and that's what it looks like.

YCbCr 4:2:2 say, is not designed to be reconstructed back to up 4:4:4, but to be dealt with and displayed as is. Sure, we can develop algorithms to take it back up to a seemingly plausible 4:4:4 (guess who sells plugins to do that!) but RAW mosaic data was designed for full RGB reconstruction, as in the Bayer pattern and RAW data are designed in tune with the eventual reconstruction to RGB, as a fully cohesive system.

 

[Bruce Allen]

Just one more thing:

It is precisely HDRI (and only HDRI) that can make digital competitive with analog film. Digital has already matched (and exceed in case of 35mm) the resolution of an image. You can also use the "same" glass and any other accessories. The only remaining mountain to scale is the high dynamic range of analog film. Without HDRI we will never catch up...

Agreed. It's just that I'd be happy with Medium Dynamic Range, let alone High. I'd currently characterize digital sensors as Low Dynamic Range...

Yes when You are talking about YUV (more precisely Y'CbCr). In RGB world 4:4:4 refers to 8-bit RGB and 8:8:8 refers to 16-bit RGB - just ask anyone with Sony Vialta film scanner. I was part of the team who pull the 8:8:8 data out of it...

So you mean x:x:x means one thing for Y'CbCr and something else for RGB? What a mess!

Looks to me like Sony has just twisted the terminology for marketing purposes.

If so, it's a horrible misuse of the : symbol, which used to imply a ratio.

Peter, let us know more about your HDRI design. What I've seen so far always involves compromises on overall sensitivity / price / feasability, etc. I'm sure it can be done in the future, just we are not quite there yet. We've had many conversations about this in the past.

To go back to the original point, how do you expect Scarlet to capture "native 4:4:4 RGB" with a Bayer sensor? Or are you talking about something other than a Bayer sensor (which we have also discussed many times)?

Again, I don't think Red One captures something equivalent to "native 4:4:4 RGB" at 4K - but when you scale it down to 2K you can make 4:4:4 quality images. Not a complex concept.

Bruce Allen
www.boacinema.com

 

[Bruce Allen]

Chroma sub-sampling notation is a measure of by how much you reduce the resolution of the chroma components when the image is in YCbCr form. "least 16-bit (8:8:8)? ", chroma sub-sampling notation does not refer, in any way, to bit depth, so that question is meaningless. I see your comment that a scanner uses such notation through for 16bit transfer, but quite frankly, that's a pure abuse of the notation and not valid in any way. Remember when HD was referred to as 22:11:11 because of the higher sampling base frequency than SD?

Yaay, Graeme agrees.

Bruce Allen
www.boacinema.com

 

[Graeme Nattress]

Bruce, for marketing, you can call anything whatever you like. Doesn't make it so though, does it? But the simple answer is, according to definition, the numbers x:y:z in chroma sub-sampling notation refer to chroma sub-sampling ratios (mucked around with so that "0" has a special meaning in 4:2:0) and refer, in no way to bit depth. HDCAM SR can be 4:4:4 in 10bit but we don't call that 5:5:5 !!!!

Graeme

 

[Bruce Allen]

Bruce, for marketing, you can call anything whatever you like. Doesn't make it so though, does it? But the simple answer is, according to definition, the numbers x:y:z in chroma sub-sampling notation refer to chroma sub-sampling ratios (mucked around with so that "0" has a special meaning in 4:2:0) and refer, in no way to bit depth. HDCAM SR can be 4:4:4 in 10bit but we don't call that 5:5:5 !!!!

Graeme

Haha, yeah, Sony is funny. The figured out that people would understand HDCAM SR > HDCAM because 4:4:4 > 3:1:1...

"...but now how are we going to convince people to buy new stuff again??!?"

Ahh "8:8:8"...

Next April 1st you guys should update your website to give Red One a spec of some kind of weird number like 52:64:17 and watch c.com go nuts!

BTW, when are you ever going to release your Nattress Effects plugins for After Effects so I can use them more? I'm sure you have so much free time...

Bruce Allen
www.boacinema.com

 

[Peter Majtan]

Chroma sub-sampling notation is a measure of by how much you reduce the resolution of the chroma components when the image is in YCbCr form. "least 16-bit (8:8:8)? ", chroma sub-sampling notation does not refer, in any way, to bit depth, so that question is meaningless. I see your comment that a scanner uses such notation through for 16bit transfer, but quite frankly, that's a pure abuse of the notation and not valid in any way.

I agree with You that the use of 8:8:8 for 16-bit is abuse, but not necessarily unjustified. As You have said Yourself 4:4:4 refers to sampling (and sub-sampling) of the data when interpreted to YCbCr form. In RGB format - there is no sub-sampling. And if 4:4:4 RGB represents RGB data in 8-bit format and then we have "double of the bits" to describe the 16-bit sampling - it is "almost" justified to use the 8:8:8 nomenclature.

Notation should evolve with the technology (just as PC's are switching to 64-bit world from 32-bit). How else would You "fairly" describe the difference between sampling the sensor data at 8-bit and 16-bit?

In no way I am trying to defend SONY - I do almost "hate" the company, especially after working with them. It is companies like them who hold us back in a stone age where is comes to digital...

 

[Graeme Nattress]

Free time, what's that? With baby and all, we're busy here.

Graeme

 

[Graeme Nattress]

Well, as we have 8bit YCbCr 4:2:0, 4:2:2, 10bit 4:2:2, 4:4:4, and we have 8 bit RGB, 10bit RGB, 12 bit RGB and 16 bit RGB, I'd use the phrase "16 bit RGB t o describe 16 bit RGB, not not use a chroma subsampling notation at all. It sounds like what the confusion is, is the use of a double dual-link HD-SDI to transfer the image? I don't think using chroma sub-sampling notation for non YCbCr data as making too much sense.

Graeme

 

[Joel Kaye]

First, I'd argue against 4:2:2 giving better colour (deeper and nuanced) than HDV (which is 4:2:0) because contrary to popular usage of the term, 4:2:2 is NOT a colour space. A colour space is something like Adobe RGB 1998, or REC709, or sRGB etc. 4:2:2 is chroma sub-sampling notation.

Probably explains why I could get better picutres out of my HD-100 than my HVX when I owned both of them. Not to mention the superior dynamic range of the HD-100.

Pretty much why I toss every number I hear now in the river and just look at the footage.

 

[Bruce Allen]

I toss every number I hear now in the river and just look at the footage.

That's a good quote.

Bruce Allen
www.boacinema.com

 

[Peter Majtan]

Peter, let us know more about your HDRI design. What I've seen so far always involves compromises on overall sensitivity / price / feasability, etc. I'm sure it can be done in the future, just we are not quite there yet. We've had many conversations about this in the past.

I would love to, but this will most likely require its own thread. Also I believe in sharing of knowledge and ideas for the greater benefit of the industry as whole - therefore I will need to first speak with Jim (are You listening?) or someone else at RED as to what will they do with such a information. What I don't want is that RED will patent the concept and prohibit any other manufacturers (who will survive the onslaught of Scarlet - providing RED is ready for it...) for implementing it in their solutions...

You don't have to wait for future - we were able to do this 2 years ago (ARRI was also working on HDRI solution, but I can't talk details about their work...).

And in no way it means any compromise what so ever - the solution is much simpler that You may think. The darkest spot is under the candle...

and (my company's motto):

You don't need eyes to see - You need a vision!

To go back to the original point, how do you expect Scarlet to capture "native 4:4:4 RGB" with a Bayer sensor? Or are you talking about something other than a Bayer sensor (which we have also discussed many times)?

It really does not matter what pattern the sensor is using (altho majority is Bayer), or how many sensors there are (it will have of course impact on cost, quality, etc... - and not always in way You may think...).

The only hint I can give so far is that it can't be CCD - it has to be CMOS. The difference is in the timing of data read-out from the sensor. In the lemon's terms CCD collects photons for a period of time (charge) and then reads-out the value (hence the limitation on FPS), while CMOS is "constantly" exposed to the photons and it is up to the electronics how fast and at what frequency it can read-out the values...

You should be able to start getting the picture...

Anyway - gotta go to an event - will be back Sunday night...

Cheers

 

[Peter Majtan]

Well, as we have 8bit YCbCr 4:2:0, 4:2:2, 10bit 4:2:2, 4:4:4, and we have 8 bit RGB, 10bit RGB, 12 bit RGB and 16 bit RGB, I'd use the phrase "16 bit RGB t o describe 16 bit RGB, not not use a chroma subsampling notation at all. It sounds like what the confusion is, is the use of a double dual-link HD-SDI to transfer the image? I don't think using chroma sub-sampling notation for non YCbCr data as making too much sense.

Graeme

I think You hit the nail on the head.

The point is that the industry in general (especially post) is now using 4:4:4 nomenclature for RGB color space (as pro-post is done in RGB), rather then YCbCr. This might be for reason that is it much easier to explain to a client in "lemons terms" the difference between 4:2:2 YCbCr (the format most clients know - but don't understand what it actually means...) format and pure RGB - which is not subsampled - hence 4:4:4. According to the same logic one could then interpret 16-bit RGB as 8:8:8 as it is not subsampled and has double the bit depth...

In which case Your notion of 10-bit RGB being 5:5:5 is interestingly "correct", and so is 12-bit RGB 6:6:6, 14-bit RGB 7:7:7 and 16-bit RGB 8:8:8...

I am not saying that this is some kind of official standard. What I am saying is that this is how in reality postproduction facilities (and many other folks in the industry) refers to it...

 

[Peter Majtan]

Pretty much why I toss every number I hear now in the river and just look at the footage.

You are right and wrong at the same time...

Of course it is up to the perception of the viewer to like or dislike the footage - and I am not even talking about the artistic value. But in the same way there are people (really!) who prefer analog audio of tape and vinyl (with its "analog" noise "richness") to the "digital pristine" of CD, SA-CD, etc...

From the technical perspective the numbers matters a lot. Especially when You start talking DI (Digital Intermediate) and postproduction. The only problem here is (as Graeme correctly pointed out) when some companies (ESPECIALLY SONY!!!) misuse the numbers for marketing purposes - but that is another story...

ADDED:

This also reflects in Graeme mentioning that "The basic decode of colour on the RED ONE is very under saturated. That's because it captures a very wide gamut of colours, one of the benefits of our RAW approach. This is a good thing!"

When You see well exposed wide gamut image - it actually looks little dull, but just wait when You see what can be done to the footage in the post!!!
On the contrary - SONY has been expanding (and incidentally clipping) the dynamic range of their cameras to boost contrast, making the footage appearing "richer" to the eyes (as compared for example with CANON), but take that footage to post and see the colorist pulling his hair out...

So, to conclude my opinion, just because something look "nicer" or "better" to Your eyes (being a very subjective terms in a first place) - it not always translates to good quality footage. Just ask any post guys or colorists...