2K RGB vs. HDCAM-SR 4:4:4

[David Mullen ASC]

So today I had to go to a post house to look at some footage from a movie I shot where the production wants me to consider doing an HDCAM-SR D.I. instead of a 2K D.I.

 

We scanned the shots on a Spirit 2K, recorded to data (2K RGB) and to HDCAM-SR (1080P 4:4:4), and then did some split-screens in the D.I. suite.

 

At first glance, the resolution was rather similar. So I zoomed into the image to get a better look at what was going on.

 

First off, the 2K scan came up much closer to correct just by putting the scan through a LUT, whereas the 1080P recording was very hard to match exactly in terms of gamma, even though both were technically 10-bit LOG images.

 

There was a layer of blue noise in the shadows on the HDCAM-SR version that was absent in the 2K version.

 

Bright highlights were clipped faster on the HDCAM-SR and there was a video-ish edge to the clipping, whereas the 2K scan was flatter in the bright highlights with more information. We had a shot with a bright shaft of sunlight coming through a window and glinting off of a shiny chair, and there was more information in the whites of the 2K scan and they did not have a certain noisy & clipped look that the HDCAM-SR version had.

 

Basically the 2K scan looked more like a film original whereas the HDCAM-SR version had some sort of noise and video-ish edge to things, more pixellated in a way. Also, the fleshtones were harder to get back to normal in the HDCAM-SR version -- they had an odd pink-grey flatness, hard to describe.

 

Now the D.I. facility said that maybe they could work further on the HDCAM-SR to "optimize" it, but it struck me that the 2K scan just came up a lot closer to correct without much work in comparison, so it disturbs me a little that with HDCAM-SR, I may be dealing with more fiddling with the signal.

 

Now they said that even though the HDCAM-SR was 10-bit LOG, it was still REC 709 color-space, which seemed odd to me, you'd think for a D.I. you wouldn't need to work in REC 709. Maybe that accounts for why the 10-bit LOG image had less highlight information in the HDCAM-SR recording than the 2K RGB version?

[John Brawley]

Now they said that even though the HDCAM-SR was 10-bit LOG, it was still REC 709 color-space, which seemed odd to me, you'd think for a D.I. you wouldn't need to work in REC 709. Maybe that accounts for why the 10-bit LOG image had less highlight information in the HDCAM-SR recording than the 2K RGB version?

 

I gather it's for a film finish david ? Were you looking just in the DI suite or at film tests ?

 

The thing that I have only realised lately is that LOG and LIN aren't colour space. REC 709 is a colourspace, but there are less values that can be stored within that colourspace matrix, no matter if the image is LOG or LIN encoded. Isn't there a wider gamut of colours in a 2K scan ?

 

When i last tested for this....i dodn't think there was much difference in terms of resolution. But it was in in the colour information !

 

jb

[David Mullen ASC]

Yes, it was for a film finish but we just looked at it digitally projected. Though I think the people who wanted me to say that there was no difference between 2K and HDCAM-SR were not happy that I watched the images zoomed in 100% at times to check for artifacts, my point was that I wanted to really know what the differences were at the grain and edge level where the detail actually exists. But even at normal size, we could see the issue with the white areas. In fact, I saw similar problems in that Kodak Super-16 demo for 7219 that was finished in an HDCAM-SR in South Africa, this sort of "sizzling whites" look.

 

I mean, it should tell you something that the HDCAM-SR version is much less than half the data rate of the 2K RGB version -- clearly some information is being lost in some form.

 

Also, isn't 4:4:4 HDCAM-SR limited to 440 Mbps in current post studio decks? So that's a 4:1 compression I think, though that's not severe. And REC 709 must also be truncating some information to fit into that color space.

[Dan Goulder]

If you go the HDCAM SR D.I. route, you may prefer the overall result by doing your primary correction prior to hitting the tape (thus doing a linear transfer), as opposed to doing all correction after the transfer, as you would normally do with 2K DPX files.

[David Mullen ASC]

If you go the HDCAM SR D.I. route, you may prefer the overall result by doing your primary correction prior to hitting the tape (thus doing a linear transfer), as opposed to doing all correction after the transfer, as you would normally do with 2K DPX files.

 

Trouble is that for a feature, you have to first conform the movie to match the offline edit before you color-correct it, so if you are doing an HDCAM-SR D.I., you are probably transferring to HDCAM-SR tapes and then using those to conform before doing a tape to tape color-correction. Otherwise, you might as well do a 2K D.I. if you are just going to scan and store on a mainframe and conform that, color-correct, and THEN transfer to HDCAM-SR -- I mean, what's the point then to going to HD tape? You might as well stay in data mode all the way to the film recording stage.

[Dan Goulder]

Trouble is that for a feature, you have to first conform the movie to match the offline edit before you color-correct it, so if you are doing an HDCAM-SR D.I., you are probably transferring to HDCAM-SR tapes and then using those to conform before doing a tape to tape color-correction. Otherwise, you might as well do a 2K D.I. if you are just going to scan and store on a mainframe and conform that, color-correct, and THEN transfer to HDCAM-SR -- I mean, what's the point then to going to HD tape? You might as well stay in data mode all the way to the film recording stage.

Normally, this approach would be initiated at the dailies stage, where everything is transferred best light to HDCAM SR, which is later used as the online master for HD or optional film out. (The 2K D.I. option can still be utilized.) This is a standard workflow now being offered at Laser Pacific. We did a project over at E-film a couple of years ago using this same workflow, and the folks at E-film were quite stoked at the results. It was probably wrong of the post house you did the tests at to say there should be NO difference in quality. However, properly done, the results should at least be close.

[David Mullen ASC]

It was probably wrong of the post house you did the tests at to say there should be NO difference in quality. However, properly done, the results should at least be close.

 

They were close, it's just that the 2K was better all around. And it wasn't the post house pushing HDCAM-SR over 2K.

 

In your EFILM tests, did you zoom into the images and compare in a split-screen, wiping back & forth? Because it's hard to see the difference in just projecting tests at normal speed at normal lab screen size.

 

In theory, they should be close, but also in theory, 2K should be better. And from what I saw, it was.

 

I actually did an HDCAM-SR D.I. before, on "Shadowboxer", with the final film-out by EFILM, a couple of years ago. That was somewhat disappointing considering I shot that movie in 35mm anamorphic. Personally, I don't want to have to do another HDCAM-SR D.I. unless the budget is really low, or if I had shot Super-16.

[Scott Fritzshall]

Yes, it was for a film finish but we just looked at it digitally projected. Though I think the people who wanted me to say that there was no difference between 2K and HDCAM-SR were not happy that I watched the images zoomed in 100% at times to check for artifacts, my point was that I wanted to really know what the differences were at the grain and edge level where the detail actually exists. But even at normal size, we could see the issue with the white areas. In fact, I saw similar problems in that Kodak Super-16 demo for 7219 that was finished in an HDCAM-SR in South Africa, this sort of "sizzling whites" look.

 

I mean, it should tell you something that the HDCAM-SR version is much less than half the data rate of the 2K RGB version -- clearly some information is being lost in some form.

 

Also, isn't 4:4:4 HDCAM-SR limited to 440 Mbps in current post studio decks? So that's a 4:1 compression I think, though that's not severe. And REC 709 must also be truncating some information to fit into that color space.

I recall that it was 440 for 4:2:2 and 880Mbps for 4:4:4, but I think the 4:1 compression is correct. I've never heard this complaint before, I kind of wonder if it's just something relating to this particular facility's setup? I'll admit to not knowing a whole lot about the inner workings of its compression (DCT, I think?) but I was under the impression that it would just be compressing spatially rather than really tossing colorspace information. Everything I've heard about SR's compression is that it's just about as minor a hit as one can get, visually-speaking, so it's kind of surprising that it's resulting in that kind of color loss.

 

How much of a cost difference is SR vs. data? You're transferring on the same machine, and doing the DI on the same machine, so all I can imagine is that the facility is charging you a lot more for data just because they can, unless there's some big piece I'm forgetting. I mean data-wise you don't even need to pay for deck rental, just the cost of the hard drives.

[David Mullen ASC]

I recall that it was 440 for 4:2:2 and 880Mbps for 4:4:4, but I think the 4:1 compression is correct.

 

Until the new HDCAM-SR studio deck comes out, I think only the SRW1 can record at 880 Mbps at 4:4:4, which I think is double-speed. But then, if that's true, how do the post house handle HDCAM-SR recordings at 880 Mbps?

[Phil Rhodes]

I could try to dream up reasons why any of that might have happened, but unfortunately my high-end post experience is becoming a little out of date. I'm not hugely surprised at your results, though. In general terms, a hard disk is a bit bucket which will dumbly store whatever you throw at it (which is the way we like it). An HDCAM-SR deck is always trying to be a Rec.709 legal device and will make assumptions about what your aims are, which is of course the last thing we want. I presume the noisy shadows are an artifact of the slight compression - the MPEG-4 quantiser being thrown off by HF detail - and it would be a typical artifact - although at such a bitrate I would be surprised if it were atrociously bad.

 

It seems to me that it would be particularly bad at storing log signals since so much of the highlight information is in the top few code values, which the deck may decide are out of bounds. Do these very high end VTRs not have a "bit bucket" menu option to turn off all the 709 bracketing? It oughtn't to be rocket science to implement.

 

This is why I am extremely wary of compression, where it's low or high, regardless of what it looks like subjectively.

 

Phil

[Stephen Williams]

Until the new HDCAM-SR studio deck comes out, I think only the SRW1 can record at 880 Mbps at 4:4:4, which I think is double-speed. But then, if that's true, how do the post house handle HDCAM-SR recordings at 880 Mbps?

 

Hi David,

 

The SRW1 will record @880, then a SRW1 must be used to play the tape back at the post house. I have done this many times. The post house has to rent a HDCAM SR player in any case it's not an issue.

 

Stephen

[John Brawley]

Also, isn't 4:4:4 HDCAM-SR limited to 440 Mbps in current post studio decks? So that's a 4:1 compression I think, though that's not severe. And REC 709 must also be truncating some information to fit into that color space.

 

 

Correct. they are about to do a new deck, the 5800 off the top of my head...that will play back the 4:4:4 880Mbs tapes. but it won't playback regular vanilla HDCAM nor digibeta like the 5500 does.

 

I think the compression is the same, it's just the colour sampling that changes. I think this is the shortfall, along with 709. It's all about money though right ?

 

jb

[Phil Rhodes]

> I think the compression is the same, it's just the colour sampling that changes.

 

When it goes from 4:4:4 to 4:2:2, does it also go from RGB to YUV? Or is it always YUV?

 

I had assumed Mr. Mullen's tests were on an SRW1 at 880mbps; if they were on plain vanilla SR, then the loss could easily have been in the colorspace conversion.

 

Phil

[Stephen Williams]

> I think the compression is the same, it's just the colour sampling that changes.

 

When it goes from 4:4:4 to 4:2:2, does it also go from RGB to YUV? Or is it always YUV?

 

I had assumed Mr. Mullen's tests were on an SRW1 at 880mbps; if they were on plain vanilla SR, then the loss could easily have been in the colorspace conversion.

 

Phil

 

Hi Phil,

 

The SWR1 will record 4:4:4 at both speeds. Just more compressed @440.

 

Stephen

[Mitch Gross]

I would say that all of David's findings are as I would expect and there ould still be a similar difference at 880.

[David Mullen ASC]

It's hard to describe since the differences were subtle, but when we zoomed in 100% and split-screened back and forth, the HDCAM-SR version had more of an electronic "sizzle" and mild artificial edginess to it compared to the 2K version.

[Stephen Williams]

It's hard to describe since the differences were subtle, but when we zoomed in 100% and split-screened back and forth, the HDCAM-SR version had more of an electronic "sizzle" and mild artificial edginess to it compared to the 2K version.

 

Hi David,

 

I preferred scans from an Arri Scanner v Spirit, they have a greater dynamic range as there is a 2 exposure capture, one very dark for the highlights & one normal. I also think the Spirit adds noise.

 

Stephen

[Michael Most]

Trouble is that for a feature, you have to first conform the movie to match the offline edit before you color-correct it, so if you are doing an HDCAM-SR D.I., you are probably transferring to HDCAM-SR tapes and then using those to conform before doing a tape to tape color-correction. Otherwise, you might as well do a 2K D.I. if you are just going to scan and store on a mainframe and conform that, color-correct, and THEN transfer to HDCAM-SR -- I mean, what's the point then to going to HD tape? You might as well stay in data mode all the way to the film recording stage.

 

For some facilities (not ours, BTW), one major point is that the conform can basically be done with video editing equipment rather than a DI system. This keeps it in a less expensive environment for a longer period. It also allows for simple and reliable storage without needing terabytes worth of drives, and with real time retrieval, unlike data tape formats such as LTO3.

 

Of course, there are ways to do 4:4:4 DI work without incurring any compression of any type, but it requires some thought as to how to do that.

[Timo Klages]

When it goes from 4:4:4 to 4:2:2, does it also go from RGB to YUV? Or is it always YUV?

 

Phil

 

yes, it goes from RGB (4:4:4) to YCrCb (4:2:2 in this case). afaik, 4:4:4 is always RGB altough the Digital Factbook by Quantel says, 4:4:4 can also be a ratio for the luminance and color difference components (Y, B-Y,R-Y)...

 

 

greets,

timo

[Jamie Metzger]

I recall that it was 440 for 4:2:2 and 880Mbps for 4:4:4, but I think the 4:1 compression is correct. I've never heard this complaint before, I kind of wonder if it's just something relating to this particular facility's setup? I'll admit to not knowing a whole lot about the inner workings of its compression (DCT, I think?) but I was under the impression that it would just be compressing spatially rather than really tossing colorspace information. Everything I've heard about SR's compression is that it's just about as minor a hit as one can get, visually-speaking, so it's kind of surprising that it's resulting in that kind of color loss.

 

How much of a cost difference is SR vs. data? You're transferring on the same machine, and doing the DI on the same machine, so all I can imagine is that the facility is charging you a lot more for data just because they can, unless there's some big piece I'm forgetting. I mean data-wise you don't even need to pay for deck rental, just the cost of the hard drives.

Spypost in San Francisco charges a base rate of $900, then $6 a minute for transferring. That is student rates as well. I don't understand why it's so expensive, but it is.

[Guest tylerhawes]

Hi David,

 

I think something must be wrong in either the transfer or the screening setup.

 

On Superman Returns I did some extensive tests for the benefit of everyone involved where I recorded DPX frames onto SR and then re-captured, then "pixel sniff" the image with an eye-dropper tool in Autodesk Combustion to see if the values changed. For most pixels in the image the values were identical, but it was not uncommon to find some 1 value variations. 2 value variations were more rare. In any case, while not actually perfect, I think everyone (including Mark Stetson and Scott Anderson if I may say so) agreed it was "visually lossless", and if it weren't for the eye dropper tool, we'd never have known the difference.

 

Besides this and anecdotally, I've supervised DI on over twenty features that have used HDCAM-SR as an indermediate format and never experienced a visually noticeable difference to uncompressed. This includes use of log and video images, and has stood-up to multi-generation such as log footage on SR, to disk, apply linearization LUT, back to SR, etc.

 

It's possible there was something specific to the deck you were using, but it's more likely in the screening room or the transfer process. Can you be absolutely sure the same bits went to the SR as to the disk? I know with the Thomson Spirit, if they don't have the new 2K model, it can be a little tricky to get log working correctly, and isn't in the original design for the machine. I don't know the device well enough to say for sure, but is it possible that the Spirit you used puts the data output through a different chain of processing than the video? I've been in situations where the video chain had a DaVinci on it while the data bypassed DaVinci straight to disk, and seen the DaVinci settings mess with the video output...

 

If that's not it, it leaves the screening setup. That's going to be specific to the room, and there's too many possibilities to speculate without knowing the configuration. I might start with where/how the LUTs are applied if you're confident in the source tape (transfer).

 

One other thing, HDCAM-SR is not 709 colorspace limited. SR doesn't have a specific colorspace, other than the constraints of a 10-bit RGB recording (OK, actually 3-1019 to be 100% accurate). So if someone told you that the 709 colorspace of SR is maybe the difference, it isn't. Colorspace, in the context of how I think you meant it, is simply a matter of intent when on SR, just like data. Whatever colorspace you intend for the images to be used in, and they are calibrated for, that is what colorspace the SR tape is. I've asked Sony to consider adding metadata to the SR tape where you can flag colorspace, but I'm not holding my breath.

 

Someone else said it, but to confirm, the existing studio decks only work in 440mbit per second. This is the same whether you're in 444 or 422. I've heard that Michael Mann even preferred 422 for some shots because it is less compressed than 444 at the same datarate, although I don't know if that's true for sure. There is a point to it, though...

[David Mullen ASC]

Well, I have no doubt that with some monkeying around, the facility could have gotten better results from HDCAM-SR, but that's the problem -- I had no guarantee that they could fix the problem, whereas they had little problem with the 2K scan.

 

See, the issue was "hey we have to transfer the movie quickly to HD for a Sundance screening, why don't we just transfer it to HDCAM-SR and skip the 2K transfer for the D.I. later?" Apparently it was such a rush job that even the difference in speed between a transfer to 2K and conversion to HD vs. a transfer to HD would mean missing the deadline, so they wanted a quick decision.

 

So suggestions that "with some more work we maybe could get the HDCAM-SR to be a closer match to the 2K scan" was too scary to contemplate because either they got it right or they didn't, right then. Because once we did all the transfers to 4:4:4 HDCAM-SR, there was a high probability that the producers were going to live with it no matter how much worse it was than 2K.

 

It seemed safer, under those deadline pressures, just to concentrate on the 4:2:2 HD version needed for the Sundance screening and not lock ourselves in yet for the entire D.I.

 

I can't explain the differences, and neither could the post house really -- all I can say is that they were there and the director and I were asked to pass judgement on them. Personally, I agree that you'd think they would be closer in quality than they were (and I was surprised when they said that HDCAM-SR was always REC 709, because even I know that's not true), but my impression was that working with a 2K RGB scan was in essense, a simpler path for the post house, whereas working with HDCAM-SR required "tweaking" and we had no time for that, no time to try other facilities, etc. That's the problem with being rushed.