4k to 1080p results in 4.2.2.?

I read somewhere that the reason 4k downconverted to 1080p looks better (if that is in fact true), is because it then becomes 4.2.2. instead of 4.2.0.
Is there any validity in that statement, and if there is, does that mean that when you colorgrade 4k on a 1080p timeline in resolve, you really get the 4x amount of colors to work with? I understood 4.2.0 means 250.000.000 colors for grading and 4.2.2 means 1.000.000.000 colors for grading?

? Within resolve any file is treated as 32 bit float.

Until you hit save, there is no difference.

A down-conversion can in some instances provide more color information than was available in the original acquisition format. But if you're expecting an obvious improvement, compared to the 4K original, you'll probably be disappointed.

Try it yourself: you should be able to find lots 4:2:0 4k footage for download.

Wouter, you got things mixed up a bit:

It's not about the number of colors, but about their resolution. With 4:2:0 colors are resolved at one quarter of the luma channel (for simplification, you may say black and white). If you have UHD, that would mean color is resolving not much more than SD (precisely 960 by 540)! If you now scale down from UHD to HD (and I expect Resolve to preserve all what it get's to work with), you'll end up with half the resolution for colors vs luma. So, yes, it should be good enough for output in 4:2:2.

Number of colors, or rather differentiation of shades of colors, is dictated by the bit depth, like 8, 10 or even 12 bit. These are technically separate values and might be used in combination, but usually 4:2:0 is stored with 8 bit. Finally, all of that is massively compressed. That's why it is very critical in grading. BTW (since we tend to think in decimals, but this is binary), you need to understand that 10 bit is not 20% more than 8, but 300%.

? Within resolve any file is treated as 32 bit float.

Until you hit save, there is no difference.

Thanks for the answer .

So for the grading is makes no difference.

Is there any validity to the claim that shooting in 4k and rendering in 1080p looks better compared to shooting and rendering in 1080?

Wouter, you got things mixed up a bit:

It's not about the number of colors, but about their resolution. With 4:2:0 colors are resolved at one quarter of the luma channel (for simplification, you may say black and white). If you have UHD, that would mean color is resolving not much more than SD (precisely 960 by 540)! If you now scale down from UHD to HD (and I expect Resolve to preserve all what it get's to work with), you'll end up with half the resolution for colors vs luma. So, yes, it should be good enough for output in 4:2:2.

Number of colors, or rather differentiation of shades of colors, is dictated by the bit depth, like 8, 10 or even 12 bit. These are technically separate values and might be used in combination, but usually 4:2:0 is stored with 8 bit. Finally, all of that is massively compressed. That's why it is very critical in grading. BTW (since we tend to think in decimals, but this is binary), you need to understand that 10 bit is not 20% more than 8, but 300%.

Thanks, I will hVe to read this a few times to really undertand it

It has to do with the resolution of the chrominance in the U and V color channels at the different sizes, as well as what chroma subsampling strategy you’re using for the different formats.

If you have a 3840x2160 4:2:2 source, you can transcode that down to a 4:4:4 1920x1080 frame size and have actual 4:4:4 values for each pixel. The math just works.

It’s not the exact same quality as if you had actually captured 4:4:4 footage in-camera, because you’re dependent on whatever software is doing your scaling, but the footage will certainly behave as 4:4:4.

Here’s a rough explanation from when I asked Charles Poynton about it. https://twitter.com/sethgoldin/status/7 ... 14144?s=21

For more background, I recommend Digital Video and HD. https://www.amazon.com/Digital-Video-HD ... 0123919266




Sent from my iPhone using Tapatalk

Just 2 articles about this subject.

In German https://www.slashcam.de/artikel/Grundlagen/4K---Von-RAW-bis-4-2-0.html

In English https://www.cinema5d.com/best-way-downsample-4k-smartphone-footage/

And 8k bayer-RAW to 4k de-bayer RGB looks freaking amazing.

It has to do with the resolution of the chrominance in the U and V color channels at the different sizes, as well as what chroma subsampling strategy you’re using for the different formats.

If you have a 3840x2160 4:2:2 source, you can transcode that down to a 4:4:4 1920x1080 frame size and have actual 4:4:4 values for each pixel. The math just works.

It’s not the exact same quality as if you had actually captured 4:4:4 footage in-camera, because you’re dependent on whatever software is doing your scaling, but the footage will certainly behave as 4:4:4.

You can do this not only for 4:2:2 source, but also for 4:2:0. It requires special approach though. Resolve is not going to provide it.
You simply scale Y channel and leave UV as is (they are already just HD in case of UHD 4:2:0 source). It's actually perfect result as you using real UV values- they are not interpolated in any way. You just scale Y, but because you have oversampled source it'll actually yield good result (better than just HD 4:4:4 capture with HD size Bayer sensor).

David Newman (Cineform) once posted that UHD 420 to 1080 444 works so long as the scale is done in the RGB domain.

Yes, a Y only scale and leaving the UV alone also works but apps don't really operate that way (scaling only one component).

https://www.eoshd.com/2014/02/discovery-4k-8bit-420-panasonic-gh4-converts-1080p-10bit-444/

Yep- not a single pro tool will do it. You need to switch to special tools, like avisynth or vapoursynth. They all allow you to do "crazy" things It's good to know therm at least a bit.

This article is not very good as bit about 8bit UHD going to 10bit HD is not really true. It may end up bit better, but it's not going to make real 10bit file.

myFFmpeg(Windows) or iFFmpeg(Mac) both work.
A pro app is an app that gets the job done...

You can do this not only for 4:2:2 source, but also for 4:2:0. It requires special approach though. Resolve is not going to provide it.
You simply scale Y channel and leave UV as is (they are already just HD in case of UHD 4:2:0 source). It's actually perfect result as you using real UV values- they are not interpolated in any way.

Exactly, in addition, you increase the bit-depth of the Y-channel by 2 as the result in the Y-channel is 4x oversampled. Even the U and V channels will have a fractional bit-depth increase in case of 4:2:2 but not for 4:2:0. Of course, all this comes with the penalty of having a lower resolution.

of course, all this comes with the penalty of having a lower resolution.

Let's hope A.I. Gigapixel comes to the moving pictures sooner than later.

Have done it semi-manually, the results are amazing but it consumes a lot of time.

? Within resolve any file is treated as 32 bit float.

Until you hit save, there is no difference.

Thanks for the answer .

So for the grading is makes no difference.

Is there any validity to the claim that shooting in 4k and rendering in 1080p looks better compared to shooting and rendering in 1080?

It makes a difference if you use secondary corrections based on color.

Finally, compression plays a big role in this. Crappy overcompressed 4K (actually UHD) footage from a phone will not suddenly look great in HD, even if you use specialized software to transcode. But 4K out of a serious camera in ProRes looks better in HD than recording HD from the start, since Bayer sensors don't really resolve HD if they have only a 1920 by 1080 raster.

This article is not very good as bit about 8bit UHD going to 10bit HD is not really true. It may end up bit better, but it's not going to make real 10bit file.

Agreed. The averaging of 4 luma values only approximates native deeper capture. The chroma gains nothing. So something in between. Maybe something. maybe nothing.

The UHD 420 to HD 444 downsample is sound. IMO.

As for quality. Always too variable to speculate. Maybe something. Maybe nothing.

Native 420 media does not chroma key very well on edges. So a UHD 420 downsampled to a 1080 444 could/should do better on keyed edges. In theory at least.

This appears to argue strongly that recording in something like 4K/UHD 10bit ProRes 422 HQ (BMPCC4K) and then downsizing in Resolve to HD deliverables will lead to better Colour than downsizing in camera. Is that a fair conclusion?

I only produce 2K/HD deliverables for clients so far and I had planned to shoot ProRes 4K/UHD and downscale in camera to record HD. If downscaling in Resolve is that much better that one is effectively working with 444 data, then I need rethink my workflow.

I had lamented that ProRes 444 was not an option in camera but this looks like a reasonable way to get there anyway. The other option is raw of course, but with the new tools at our disposal ProRes can be a faster workflow.


Sent from my iPhone using Tapatalk

This appears to argue strongly that recording in something like 4K/UHD 10bit ProRes 422 HQ (BMPCC4K) and then downsizing in Resolve to HD deliverables will lead to better Colour than downsizing in camera. Is that a fair conclusion? I only produce 2K/HD deliverables for clients so far and I had planned to shoot ProRes 4K/UHD and downscale in camera to record HD. If downscaling in Resolve is that much better, then I need rethink my workflow.

You should experiment, but I would expect that if you capture in UHD 4:2:2 Y’UV and downscale in Resolve to 1080p 4:4:4 RGB, you will get better results than just capturing 4:2:2 at 1080p in-camera. Capturing 4:2:2 1080p in-camera would be needlessly throwing away color information that you could otherwise retain.


Sent from my iPhone using Tapatalk

Seth, I’ll do that test as one of the first few tests I do. I really want to try several things to see how the new camera performs. I could also do this test on the URSA Mini 4.6K say:

1) record open gate with 4.6K in raw,
2) record open gate with 4.6K in ProRes 444/XQ,
3) record open gate with 4.6K in ProRes 422 HQ and then
4) record in raw in a HD window,
5) record in ProRes 444/XQ HD downscaled from the full sensor,
6) record in ProRes 444/XQ in a HD window,
7) record in ProRes 422 HQ HD downscaled from the full sensor, and
8) record in ProRes 422 HQ from a window,

Then in Resolve compare colour on an HD Timeline
A) from the open gate and HD raw with ProRes 444 optimized media,
B) from the recorded ProRes 444, and
C) from the recorded full sensor ProRes 422 downscaled in camera and
D) the HD windowed ProRes 422.

If the results from step C look more like step A and B than the results in step D, that will prove the theorem.

I’ve done comparisons previously of every flavor of ProRes and the differences between each is fairly subtle but present; these tests will be illuminating.


Sent from my iPhone using Tapatalk

If HD ProRes XQ is recorded from much bigger area than HD (at least 1.5 x bigger) than I don't think that recording UHD 4:2:2 and downscaling will give you any real gain. In this case most optimal should be something like 2.8K RAW, which should give you "perfect" HD and in the same time save on space and processing power needs. Shame BM doesn't do it.

That's right, because of this Red offers 3K.

If HD ProRes XQ is recorded from much bigger area than HD (at least 1.5 x bigger) than I don't think that recording UHD 4:2:2 and downscaling will give you any real gain.

if you really want to eliminate the subsampling artifacts, you simple need alt least UHD for this kind of refinement.

In theory. Sensor has noise, etc and I bet you'll see no meaningful difference in real world.
Also- benefit is that you start with RAW not ProRes out of camera.

When you have 20 bucks or euro's left buy this:

http://www.ffworks.net/store.html when you're on a Mac or

https://www.myffmpeg.com/store.html when you're on Windows

and do this

https://www.cinema5d.com/best-way-downsample-4k-smartphone-footage/

When you want more resolution afterwards try this