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

[Chuck Z]

Craig, that's another question altogether! It's a question that applies to all footage. What I'd suggest is to keep the quality as high as possible as long as possible, only reducing it significantly for any distribution codec.

You hit the nail on the head.

People should do their own tests to determine if their combination of the RED Mysterium-X sensor, lenses, filters, etc. warrant keeping the chroma resolution at its maximum setting through a given workflow. If you feel it's invisible to the naked eye and/or you don't have the disk space/bandwidth to cope with this problem, then by all means subsample the chroma channels.

[Karim D. Ghantous]

It's threads like this which make REDUSER such a valuable use of my time. So thanks to all!

However, it is my opinion that instead of such a simple decimation, you'd actually get a better image by keeping the chroma resolution full, and using a slightly more lossy compression on the chroma

Wait. Did you mean 'luminance' there? Anyway I don't pretend to understand this stuff in great detail. However, Bruce's opinion makes sense as a default approach on the matter (even before correction and grading are done) based on my limited experience with this kind of topic.

[Graeme Nattress]

Nope, I meant chroma. I'm suggesting a modern codec might work better by keeping the chroma resolution full and just compressing it harder, than decimating it's resolution first then compressing it less.

Graeme

[Bruce Allen]

In terms of compression the reduction from 4:4:4 to 4:2:2 reduces the pre-compression bandwidth by 1/3, and usually in a very visually transparent way, making good use of the foibles of the human visual system. However, it is my opinion that instead of such a simple decimation, you'd actually get a better image by keeping the chroma resolution full, and using a slightly more lossy compression on the chroma to make-up for that 1/3 head-start that 4:2:2 gets over 4:4:4.

Graeme, I've been waiting since 2007 for you to do exactly that with RedCode RGB. Or for Cineform to make a codec that isn't buggy on Macs.

Since neither of these have happened yet, I am going with my next best option:
OpenEXR with 4:2:2 and B44 compression.

Maybe there will be a 4K version of Avid DNxHD444 soon.

Again, I don't see much real chroma info being lost in my real-world tests. But please point me to an alternate compression method if there's one out there that can work with all of my apps?

Bruce Allen
www.boacinema.com

[Graeme Nattress]

I wish there was time in the day to come up with something there for you Bruce!

Of course, as you know there's not so much detail in chroma to begin with which is why the human eye didn't adapt to seeing it and why YCbCr 4:2:2 etc. compression systems have been used so successfully.

Graeme

[Bruce Allen]

I wish there was time in the day to come up with something there for you Bruce!

Of course, as you know there's not so much detail in chroma to begin with which is why the human eye didn't adapt to seeing it and why YCbCr 4:2:2 etc. compression systems have been used so successfully.

Graeme

Graeme, you are a very nice man.

I wish you some days off first, instead!

If I ever start a post on my 4K 4:2:2 compressed OpenEXR 48fps stereo workflow I will make sure to include an Honorary Graeme Disclaimer that:

"the moment a good compressed 4:4:4 RGB codec becomes available, ditch this chroma-subsampling nonsense! It makes no sense in the digital world! But here's my kludge for now..."

Cheers :)

Bruce Allen
www.boacinema.com

[Tom.Wong]

Epic R3ds are 16 bit

what about the A/D conversion? is the sensor natively 14 bit going into a 16 container? 16 to 16? 14 to 12 than padded in a 16 bit container? r1 mx was a 12 bit codec, I'm wondering if the A/D stayed the same off the r1 to the Epic and if it's just 12 bits spread out in a 16 bit container.

[Craig Parkes]

I admit my question was a bit cheeky, because I knew it was an altogether different question, but I also knew it would get the sort of responses I was looking for, which is that:

A) Theoretically modern RGB compression could do a better job than 4:2:2 downsampling pre compression, but hell, 4:2:2 is fast, known and visually pretty hard to spot so heck, so rather than bemoan the world we live in going with that for now makes sense.

and

B) Bruce might write up his 4:2:2 Open EXR workflow in a post, which is a workflow that makes a heck of a lot of sense in my head theoretically, but isn't one that would be considered the 'norm' in tradition heavy iron type workflows where a lot of the literature comes from (especially not at 4K resolution.) and one that I would need to do a bit of research to implement. Bruce, if you are ever in Auckland, New Zealand, let me buy you a beer. :)

Having only really had my head in the post game for about 5 years, and having spent much of that scrounging around for scraps of information on prosumer centric forums or boffiny personal websites of video engineers to get to grips with a lot of these concepts (because somewhat surprisingly at first I discovered even a lot of people who worked day to day in industry really had no idea about what was actually going on under the hood and would spout misinformation at the drop of a hat.) I am truly thankful for the quality of communication and openness with information that is on Reduser!

[conrad gaunt]

The reason, of course, why such analogue era compression as 4:2:2 works so well is that we don't see much detail in chroma compared to luma.

The issue of how colour filter array sensors see colour resolution compared to luma resolution is a different issue. In such a system we're dealing with broad colour filters where any detail is visible in all three channels in most circumstances.

Graeme

Raw also reduces the resolution of the blue and red channels (effectively the chroma, or parts that are less important for luma), to make best use of the available total resolution of the sensor. The ratio 2(G):1(R):1(B) is exactly the same ratio as 4:2:2..

The reasons may be different, and your debayer may extend the luma further.. but lets not beat about the bush, bayer is not 4:4:4, its much nearer 4:2:2..

Of course the colour filters aren't perfect at filtering out R, G or B.. and each pixel must contain some quatity of R,G and B.. and sure, 3 chip systems don't align perfectly etc etc ..but why don't you just say a 4k raw downsampled to 2k is similar to 8:4:4.. ;)

[Miltos Pilalitos]

.. but lets not beat about the bush, bayer is not 4:4:4, its much nearer 4:2:2..

I think you didn't pay attention to what Graeme said. Your statement doesn't compute.