uncompressed 4:4:4 codecs

Which uncompressed 4:4:4 codecs (for 8-bit, for 10-bit) work well with FCS 3?

Dennis Couzin
Berlin, Germany

Don't think there is such a thing. At least not in FCP. The only 4444 codecs I know of are Animation and ProRes 4444. Animation isn't a playable codec in FCP, and ProRes 4444 is Compressed...but compressed well. The UNCOMPRESSED options in FCP are 4:2:2.

What do you need this for? (Please don't say for your Canon 5D footage...)


www.shanerosseditor.com

Listen to THE EDIT BAY Podcast on iTunes
[itunes.apple.com]

Shane, the old Apple Developer document "Uncompressed Y'CbCr Video in QuickTime" describes in detail codec 'v308' (8-bit-per-component 4:4:4) and codec 'v410' (10-bit-per-component 4:4:4). I don't know if these codecs were once available, or just in-house, or just in-theory.

FCS/FCP comes with certain codecs. Is it not possible to install other codecs?

* * *

The reasons for choosing 4:4:4 over 4:2:2 are well enough known. An example involving red-on-black titles was discussed in here a few months ago.
My reason for choosing uncompressed over ProRes at this time is archival. Uncompressed video files (excepting the headers) are transparent and non-proprietary.
So uncompressed 4:4:4, which is just 50% larger than uncompressed 4:2:2, is an ideal archiving format. It'd be simpler to handle than thousands of Y'CbCr TIFF files.

Dennis Couzin
Berlin, Germany

DOn't know about those codecs. As for other codecs, I don't think you can install others and have FCP work fine with them. DNxHD, when installed, allows for editing of DNxHD Qt files, but no external video via capture cards, and no RT effects. Exporting can give odd results too. FCP is designed to be used with FCP codecs.


www.shanerosseditor.com

Listen to THE EDIT BAY Podcast on iTunes
[itunes.apple.com]

>What do you need this for? (Please don't say for your Canon 5D footage...)

That's a good point, as almost all cameras shoot to 4:2:2 or 4:2:0, and the same is true for most delivery formats, you get no quality gain by rendering them to 4:4:4.

I remember some 4:4:4 formats... Here's one:
[www.bluefish444.com]

And there is also AJA's RGB log codec (not sure if the codec does video gamma to log conversion):
[www.aja.com]

But that is unlikely to be blessed by FCP's RT engine. Few are. Also, FCP doesn't render RGB at 4:4:4. The AJA codec is 10 bits, I read, but only because of some clever hackery in that it gets FCP to "see" it as a Y'CbCr codec or something like that, so FCP can render to it in Y'CbCr in floating point (r4fl).



www.strypesinpost.com

Quote
strypes
That's a good point, as almost all cameras shoot to 4:2:2 or 4:2:0, and the same is true for most delivery formats, you get no quality gain by rendering them to 4:4:4.

Not necessarily. Red titles were an example where there is quality gain. They're generated in FCP, not by the camera, but editing often involves more than splicing images end to end.
For various effects done on 4:2:2 material in a 4:2:2 sequence the results are different from doing the same effects on the same 4:2:2 material rendered in a 4:4:4 sequence. Even something as simple as moving the image 1 pixel to the right must come out different -- cleaner -- in the 4:4:4 sequence.

Quote
strypes
I remember some 4:4:4 formats... Here's one:
[www.bluefish444.com]

This sounds promising. Thanks!

Quote
strypes
Also, FCP doesn't render RGB at 4:4:4.

What do you mean by this? When an RGB TIFF image, or an RGB codec "none" clip, is rendered to ProRes 4444 what gets lost?

Dennis Couzin
Berlin, Germany

>Also, FCP doesn't render RGB at 4:4:4.

My bad. I meant FCP doesnt preserve 10 bit RGB precision. It only renders in 8 bit RGB. You can see that in sequence settings, and 8 bit RGB is virtually useless as a mastering format especially if your final output is for broadcast, as you will have to convert back to Y'CbCr. In theory QT supports greater than 8 bit precision, but I don't see that option very much. Also, there is a QT issue which produces a gamma shift, noise and bad truncation of values. Try exporting a tiff image from FCP from an uncompressed clip and then exporting a tiff file from After Effects. The image undergoes the legacy 2.2 to 1.8 gamma conversion with lots of noise, some banding and bad smearing of chroma. Also, you do not get a 16 bit option for tiff images. So imagine all those r3d QT proxies that you export a tiff sequence with... This is why I use FCP for offline editing of RED, not for online.



www.strypesinpost.com

>Red titles were an example where there is quality gain.

Yea, but your source is 4:2:2, sometimes 4:2:0, and then your mastering format- DVD is 4:2:0, BluRay is 4:2:0, h.264 is 4:2:0, your hdcam sr master is 4:2:2. Your monitoring format is probably 4:2:2 via SDI. You gain nothing by having an intermediate master at 4:4:4, because you end up throwing those extra pixels away. FCP is able to render in 32 bit floating point at 4:4:4 Y'CbCr, but that gets encoded to whatever your sequence format is, and it is Prores for most folks.

But supposing you shot RED or Arri Alexa or film and you are going to DCP for theatrical projection... Yea, you need 4:4:4. But you offline in FCP and export an XML at the end of the day. The Smoke guy, the guy on Resolve/Baselight/Lustre, the Flame guy, the Scratch guy, the Maya folks, the After Effects team, the Nuke guys... Those do 4:4:4. FCP is not an environment where you want to juggle with 4:4:4 DPX files or do heavy debayering of RED or do your final master off a DLP projector from. You want light files with reasonable clarity. You want fast renders to show proof of concept and whisk that offline cut to the network guys. You work with lots of rushes, you don't want a 200TB high speed server with infiniband for your multicam reality show. You want a fast, stable environment to creatively tell a story in. That's the offline environment.



www.strypesinpost.com

Quote
strypes
Yea, but...

Yea, but we should not presume that videos shot in lowly formats are always delivered in lowly formats.
For example, we have a video shot in lowly 4:2:0 AVCHD 1080/50p, 4:2:0, edited as ProRes 422 1080/50p, going to a big international film festival next month. Of course they can't project 50p video. But they can project DCP, and we'll supply a 48 fps DCP. Then the small advantages of the 4:4:4 rendering will be maintained.

Also, there are reasons for using 4:4:4 codecs for editing even when the delivery format is sub-4:4:4. Another festival insisted on HDCAM for the same video. HDCAM is 3:1:1, and just 1440 pixels wide. I don't know how the Sony deck does its scaling and rejuggles the chroma subsampling. Probably it first expands the 4:2:2 to 4:4:4 using simple chroma interpolation. Then it scales the 1920 pixel width to 1440 pixels. Then it subsamples 3:1:1. The 4:4:4 expansion will have jaggy red titles while a ProRes 4444 version, requiring no expansion, wouldn't. Granted, the subsequent 3:1:1 subsampling and its eventual expansion will introduce jaggedness, but I think the one time jagginess will be less than the two time.

Consider two examples of editing "effects" on a 4:2:2 video.
Example 1: Move the 4:2:2 video 1 pixel to the right. Then move it 1 pixel to the left. The video suffers serious chroma smearing (which can become luma smearing due to the non-independence of luma and chroma).
Example 2: Render the same 4:2:2 video 4:4:4. Move it 1 pixel to the right. Then move it 1 pixel to the left. Then render it 4:2:2. The video is as it began (except for 1 lost pixel on the right edge).
From these two unrealistic examples you can imagine more realistic ones. The point is that things we do to videos can get done poorly in 4:2:2 codecs.

Dennis Couzin
Berlin, Germany

I wouldn't call HDCAM SR a lowly format.

For your first case, rendering as Prores 4444 would be helpful to the text.

For the HDCAM deliverable, I would not expect any difference between rendering in 422 or 444, as SDI (which the signal will have to go through) is 4:2:2.



www.strypesinpost.com

I didn't call HDCAM SR a lowly format, but implied that HDCAM is. HDCAM is still very popular in Europe, where HDCAM SR didn't catch on.

Thanks for the info about HDCAM decks using SDI which limits the 444 to 422. The video lab that made the HDCAM didn't tell me this. 422 is more deeply entrenched than I imagined.

Dennis Couzin
Berlin, Germany