Resolution And Image Quality In 2k Work Flow

[Nir Shelter]

In an attempt to figure out the best way to export 2k footage whiles maintaining the highest level of color depth and resolution. I experimented with the same shot in Redcine and FCP exporting using different codecs. The results are confusing.
First export:
Loaded the shot into Redcine and decided on using the first 67 frames of the shot. It is of the back of a person wearing a blue sweater the woven threads give a good degree of detail to work with. I exported the clip using a set quality of lose-less, ratio of film2K and all in Quicktime. I then exported once using Proress HQ and once using Uncompressed 10-bit.
Second Export:
Loaded the H proxy into a FCP time line and exported the same 67 frames of the same shot using the default settings that came up then a second export using Proress 1920 1080 HQ... .

I imported all the above exported QuickTime movies to the same time line. I then identically changed the same parameters to them all to make the amount of detail more obvious, I De-saturated them and increased contrast. I then exported still frames as JPGs at grayscale best quality. Then enlarged the same cropped section (or as close as I could get it with a cropping tool) in each of the frames to bring out the differences in detail.
These are the pics of these frame grabs:

Redcine export Uncompressed 10-bit:

Redcine export Proress HQ:

FCP export Proress HQ:

FCP export default settings:

The left side of the subject's back, the left shoulder and the left arm show the most difference in detail between the exports. As light and dark contrast patches.
The FCP H proxy Proress export has arguably more detail than the FCP H proxy default settings export. Yet it appears as if they both have more detail than the Redcine exports.
Why is this when the proxies are referencing the same data as the clips in Redcine?
And does this mean that for 2K work flow in regards to picture quality. It would be better to work with the H proxies in FCP and export a QuickTime directly from the time line, then out to color or a pro grade?

 

[Steve Phillipps]

Very good test, 'cos you're only changing 1 variable it makes it a good side by side comparison. I'd be very interested in knowing the answer to your question too. Presumably if you can avoid REDCine and just stay in FCP all the way you'd be happy, I know I would.
Steve

 

[M Most]

The FCP H proxy Proress export has arguably more detail than the FCP H proxy default settings export. Yet it appears as if they both have more detail than the Redcine exports.

Uhh, I don't know what you're looking at, but what I see is that the Redcine exports have much more detail in the darker areas, and the Prores versions have a bit more contrast but tons of artifacts caused by a combination of 8 bit sampling (quantizing artifacts) and compression. The contrast differences are understandable given the limits of 8 bit processing in the Final Cut versions, but don't mistake that as a positive thing. You would get a far better color corrected final result from the much cleaner Redcine exports.

 

[Alex Carr]

Agreed. Pro Res is 80-90% of an uncompressed version of the same clip. Pro res is offline only. Its the DV-NTSC for the HD Resolutions. So expect Quality Loss.

 

[Nir Shelter]

This is an experiment for an add we have coming up soon for which I want to use Redone. The add will predominantly be for analog as well as digital broadcast, HDTV and posted on the client's web site. There for 2K will be enough even if suddenly the client wants the add for a small theatrical release.
Opinions on preferable 2K work flow will be much appreciated. On that note would it yield a better result to shoot 2K or 4K and down convert?

The contrast differences are understandable given the limits of 8 bit processing in the Final Cut versions,

That's true I did use 8 bit rendering effects in FC I there for created a new 10-bit uncompressed time line imported and stacked the Redcine uncompressed and Proress one over each other with the FC H proxy imported clip.
To emphasize the detail I freeze framed the first frame and duplicated the stack 4 times then increased the size of the clips. I watched the result on a big SD monitor, the majority of the adds potential views. I sat a friend with minimal knowledge of compression, formats and bit rates in-front of the monitor as well. He seamed to prefer the H proxy clip over the others probably due to the higher contrast in it. This was true until the 1000% image. Of course he didn't take into account the latitude, loss of detail or potential color correction and so forth. He just saw more contrast which is what eyes see the difference between pixels not the number of them or what they comprise. We cant always necessarily differentiate between digital artifacts and detail until we look closer.
These are pics exported from the FC 10bit time line to JPEG, "best color depth" and "best quality". It doesn't look like the images suffered much through the JPEG compression so it still gives a fairly good representation.

100%:

400%:

1000%:

At 100% size the images look very close to each other and almost tend to look better in the H proxy FC import. Yet at 400% the compression of the H proxy file is noticed not much of a difference between the Prores and uncompressed out of Redcine. 1000% though H proxy falls apart and there is a marginal difference between the Proress and the uncompressed mostly noticeable around the folds of the wrist and under the knuckles.
Guess that on the surface Proress holds up pretty well but will probably not during color correction or compositing definitely an of-line option. Needless to say the proxy is out of the question for anything but web delivery.
Opinions?

 

[M Most]

Opinions?

Well, my first opinion is that if you're going to compare images, you have to take obvious differences out of the equation. The question isn't what the images look like right out of the box, especially when you're working with images that are available in RAW form. The question is how they compare when you make them look the way you want them to, which means that you must color correct both to your liking and get them to reasonably match. If you can't do this yourself, you should find a colorist or someone else who can do that. You should also be aware that you can add a degree of sharpening during the debayer process if you feel you still need that once you get the contrast right.

Second, I agree with your basic conclusion - the proxies are fine for editing, and for some Internet and basic video deliveries, but for a master quality product, you need to use the original files, properly debayered, in order to get the best quality, the least noise, the fewest artifacts, the most latitude for color correction and other manipulations, and the best color fidelity.

 

[Mark L. Pederson]

Second, I agree with your basic conclusion - the proxies are fine for editing, and for some Internet and basic video deliveries, but for a master quality product, you need to use the original files, properly debayered, in order to get the best quality, the least noise, the fewest artifacts, the most latitude for color correction and other manipulations, and the best color fidelity.

YES YES YES YES YES YES

I am sooooooo tired of people telling folks to "just use the 2K Proxies". You CAN - but it does not do this camera justice.

If I had one dollar for every GIG of R3D that people are screwing up in post on RED projects I'd take a nice, LONG vacation.

Sure, you need speed, you don't need FULL MASTERING quality? Okay fine, 2K Proxies.

High-end professional work - debayer or die.

 

[Nir Shelter]

Well, my first opinion is that if you're going to compare images, you have to take obvious differences out of the equation. The question isn't what the images look like right out of the box, especially when you're working with images that are available in RAW form. The question is how they compare when you make them look the way you want them to, which means that you must color correct both to your liking and get them to reasonably match.

Absolutely I am sick of people judging footage based on it's raw look. Try explaining that to a producer or add agency rep who's knowledge of video formats is limited to their latest Nokia model hand set, on location viewing dailies shocked that the image is so "flat" or my favorite "un-saturated".
All I was trying to do with this test footage (not part of any intended project) was to assess the quality delivered by each post work flow. To figure out what way of processing the footage will yield the most information to work with then after. I guess it turned out to be the obvious one, 10-bit uncompressed out of Redcine.

But still what would be the preferred shooting format for a 2K delivery, 4K for a down convert or strait up 2K?
I'm inclined to think 2K...

 

[Rob Gardner]

I think the question for me, since all of our work is shot with a either/and/or a Varicam or HDX900, and all of it is long form (so rendering time is certainly an issue) is: how does the H proxy cut in FCP and laid back to with a Kona 3 HDcam 1080p for delivery compare to Varicam 720 24p cut in FCP and upconverted with a Kona 3 and laid back to HDcam? The tests that we have done with this (the H proxies) look very good on the HD monitors that we have. But I have not done a side by side test, because I was working with a borrowed Red. Everything we do goes through a Harrison Legalizer as well, in the course of making the air masters and I have no idea if this would have an impact on the image quality in the way we are discussing here (I think not).

When our own Red is delivered (hopefully, before I die of old age) we will do comparison tests. Of course, given the inability to compare with the same lens, it will not be entirely a fair comparison, but it should allow us to see how the image compares overall. We produce specials and series for PBS and the cable systems. The major attraction of the Red for me has always been the use of 35mm cine lenses.
Rob Gardner

 

[Alex Carr]

Now that you have test files take those into Photoshop and see how much latitude you have, I would say the Uncompressed kept some of the shadows and highlights that the proxy did not.

What part of 'Proxies are Offline' dont people understand? They frickin suck to edit with for starters, and the proxies you run through compressor are better in terms of Offline maneuverability, but STILL OFFLINE

 

[Rob Gardner]

We just finished a series of 13 one-hour shows with four editors working at the same time (shot in DVCproHD). If we had tried to factor in rendering all of that footage through Redcine I am hard pressed to see how we could have made the delivery schedule.

I think it needs to be made clear what the end product is (for us, it is a digibeta tape for SD delivery and an 1080pfs HDcam for HD). And the question is: Will the proxies hold up as broad cast masters, in comparison to DVCproHD orginal up-resed to 1080? I have no doubt that the image quality would be better to go the slower route, but would it be enough better to justify the render time? To my eye, on a CRT HD monitor, the tests we did with Red proxies and going back out to HDcam looked very good, better, I think, than Varicam origination. But I have not done it side by side with the same workflow. I will when I get my Red camera, in the fall.
Rob Gardner

 

[Alex Carr]

Its about the same time, I just finished 8 5-40 minute spots on HDCAM SR, I had to capture everything as offline DV, EDit and then Online. THEN I render the sequences before output.


With Red the only thing missing is the HDCAM deck, a program called Crimson acts as your deck.

So you : vs. (Tape based workflow)

1] Drop all _H or _M proxies into Compressor (capturing DV through Kona)
2] Edit 1080 (Edit DV)
3] Online DPX through Crimson XML in Redcine (Capturing the Edited spots from HDCAM SR)
4] Color, with RED DPX, then make Uncompressed, (use uncompressed, then render anyway)
5] Print (printing)