Is 4K 8bit 4:2:0 actually 10bit 4:4:4 in a 1080 timeline?

I’ve read FCPX might be handling it as if it was (until exported). True? Can anyone confirm? Thanks!

Where did you read this? Can you provide a link?

I've seen this several places too. Some say that downscaling 4K to 1080 uses the extra color info of 4K to create better color depth in 1080. It always seemed bogus to me but it's beyond my knowledge to judge. Here's one description:

www.eoshd.com/2014/02/discovery-4k-8bit-...rts-1080p-10bit-444/

ddixon wrote: … downscaling 4K to 1080 uses the extra color info of 4K to create better color depth in 1080. It always seemed bogus to me ..

It is.
8bit is 8bit.
There's no 'extra' color-info in a 4k video.
No way to re-create 'lost color info' by doing some math, to convert 4 pixels => 1 pixel.

You can spread the 8bit-info to a 10bit-scale (or 12 or 14 ...) = that doesn't 'add' any value.
And when you finally export within the specs of any actual (consumer) delivery format, it's 8bit again. :lol:

Forthcoming h.265 supports officially 10bit ..........

Karsten Schlüter wrote: 8bit is 8bit.
There's no 'extra' color-info in a 4k video.
No way to re-create 'lost color info' by doing some math, to convert 4 pixels => 1 pixel.

This is both true and false.
Sampling down 4 to 1 pixel the right way could create a new color value which doesn't match the 8bit colorspace but fits the 10bit colorspace.

- Andreas

Andreas Kiel wrote: … Sampling down 4 to 1 pixel the right way could create a new color value which doesn't match the 8bit colorspace but fits the 10bit colorspace....

in math.
But in real-life?

If your 4k-recording records 4 pixels with
red/blue
blue/red

the math average for fullHD is calculated to a single pixel
purple

.. but IF there was purple, why didn't the 4k recording did
purple/purple
purple/purple?


That math-average, even in 10bit misses reality.... it is just an approximation to reality...

... plus: all those discussion about color-sampling, pixel resolutions, 'sharpness' etc etc etc are mostly done by engineers - who COMPLETELY miss one important point:
'recognition'

This pic




contains just a few 'bits' of info - engineer speak.
(haven't measured it, me no engineer, could be 4bit/40x50 res

But, closing your lids a bit, shows lots of 'detail', which cannot be stored inside the file with 'just' 4bit/0.0002Megapixels!!!


..but you do see 'eyes open/closed', you do see 'single hair', you do see 'details' which are not there!

:woohoo: :woohoo: :woohoo:

The timeline in final cut I believe is 32bit floating point, which means it has pretty much has an unlimited colour depth. When you color correct 8bit footage for example it doesn't 'step' . Its not colour subsampled on the timeline either so it's not 444 or 422, it's just rgb

That said a 4k image does have more information, and when you downsample it final cut does make a decision on how to do that. It doesn't just throw 3 out of 4 pixels away. It averages pixel values out, so yes you do get more colour and luma information. It's not necessarily accurate to the scene though as it's just the computers best guess.

So no. It's not 10 bit 444, but if you record more information, as you do when recording 4k, you get more information to work with on the timeline

BenB wrote: Where did you read this? Can you provide a link?

Maybe a better way to have phrased this would have been, "I’ve read NLEs might be handling it as if it was (until exported). Is this true of FCPX?"

Couldn't recall off-hand where I'd seen this, but one source has already been linked to (EOSHD). And I know I've seen others.

Am mainly interested in this in terms of comparing to transcoding to ProRes because my normal workflow would be to transcode footage to ProRes 4:2:2 (reg) on ingest. Obviously this takes up disk space, but when working on slower machines or color correcting, etc. seemed to speed things up and (theoretically, at least) possibly result in a better end result.

So, of course, the prospect of having FCPX treat 4K footage as 10bit 4:4:4 when placed in a 1080 timeline was intriguing. Especially if working on a fast system capable of smoothly handling 4K without needing to transcode.

On a related topic, there's also the issue of which flavor of ProRes is best for 4K. I do have a link to one discussion:

www.reduser.net/forum/showthread.php?937...4444-which-is-better

And the post from that which seemed to make the most sense to me was - "The way I approach it is I use 422 as a delivery CODEC and 4444 as a working CODEC, especially if I want to do any grading/fx work."

This is largely (although not entirely) a math discussion and I'm automatically disqualified because my eyes glaze over pretty quickly and my brain shorts-out when things go in that direction.

Bottom line is; does the possibility that FCPX treats 4K as 10bit 4:4:4 in a 1080 timeline preclude the need to transcode to ProRes? Assuming one is working on a fast 4K capable system. And, if that's not the case, which flavor of ProRes is recommended as a working codec when ingesting 4K 8bit 4:2:0 and putting it into a 1080 timeline in FCPX?

Thanks!

Seamusvfx wrote: The timeline in final cut I believe is 32bit floating point, which means it has pretty much has an unlimited colour depth. When you color correct 8bit footage for example it doesn't 'step' . Its not colour subsampled on the timeline either so it's not 444 or 422, it's just rgb

That said a 4k image does have more information, and when you downsample it final cut does make a decision on how to do that. It doesn't just throw 3 out of 4 pixels away. It averages pixel values out, so yes you do get more colour and luma information. It's not necessarily accurate to the scene though as it's just the computers best guess.

So no. It's not 10 bit 444, but if you record more information, as you do when recording 4k, you get more information to work with on the timeline

Thanks, Seamusvfx. Very helpful. So I guess a logical question would be just how good at guessing is FCPX?

I really should start making a habit of noting sources when running across things and taking notes. Unfortunately, I didn't note the source on this, but it seems to relate to what you're saying, in terms of luminance:

"four 8-bit luminance samples are being summed together (presumably) which would give us 1,024 brightness levels for each quad-pixel block, which would be 10 bits."

Transcoding on injest makes 0 difference to the quality. Absolutely none. The only time it's significant is when you are exporting, either to deliver or go to another piece of software. Say you did colour correction in final cut and wanted to take those corrections to after effects for additional work, then it would make sense to go to prores 4444, but while your on the timeline transcoding only makes a difference to speed and disk space , not to quality.

Fcpx is as good at guessing as most software. There standard algorithms that pretty much all software uses for down sampling. In resolve you can change the method I believe but generally it makes very little difference.

The idea of 10bit vs 8bit codecs breaks down on the timeline, so it's not really accurate or useful to say it's being converted to 10 bit. All you need to make sure of is your export codec has enough bits to contain the information you need, so a 10bit codec is not a bad choice if you want to preserve maximum quality.