Shooting in Raw

[David Heath]

Can I enquire about the methodology, Jim?

A lot depends on what the end result is destined to be - straight 1920x1080, or something which may be viewed via 4k equipment etc.

I was previously assuming the former. Whatever the workflow, the aim being to end up with a 1920x1080 final product.

If so, the comparison should be between the 1080 ProRes and the RAW output deBayered, and then downconverted to 1080. The comparison of the charts seems to imply that's not how the comparison is being done - it reads as though the 1080ProRes is being rescaled up to match the dimensions of the deBayered raster?

If so, it's hardly surprising there is a difference, but is it not one that would be largely eroded by subsequent downconversion to 1080? I'd be more interested to see the RAW footage downconverted to 1080 and put straight up against ProRes 1080.

I take your points about future asset mining in some future situation, but for those with the money it's the difference now with 1080 that most matters.

[Jim Arthurs]

Can I enquire about the methodology, Jim?

Yes, I was scaling Prores to match the frame raster of the native ARRIRAW. And rightly so... how else can we demonstrate the full extent what is lost? The internet, as I found, is not flooded with these kind of Prores vs ARRIRAW tests. :)

As I mentioned, I doubt many people, and probably people who SHOULD know, have ever seen the difference in resolution side by side. This is real, usable resolution that is disposed of by the low-pass filtering of the codec and the in-camera down-sample... rightly or wrongly depending on the project, granted, but very much absolutely lost. This loss is, in a very real way, a loss of already paid for production value.

I'd be more interested to see the RAW footage downconverted to 1080 and put straight up against ProRes 1080. <snip> ]I take your points about future asset mining in some future situation, but for those with the money it's the difference now with 1080 that most matters.

Certainly a valid request. I had not read every post in the thread, and see you point. Easy fix... Here is a layered Photoshop file so you can pixel-peep the differences. This file is untouched 1080p Prores on the top layer, and the bottom is ARRIRAW downsampled to match the 1080p raster size. No fancy down-sample methods were used beyond simply using the "better for downsampling" option in Photoshop. Toggle back and forth between the layers in Photoshop to compare...

http://ftp.datausa.com/imageshoppe/o..._RAW_1080p.zip

What I see is a significant "pop" in overall sharpness in the ARRIRAW, specifically seen in the elimination of most chroma aliasing at about 1000 LPPH. Remember, the chart is exactly 2X distant from the camera than normal, and all numbers on the chart have to be doubled. What this tells me is that I'm going from 800 usable line pairs up to 900 solid ones by shooting ARRIRAW compared to Prores. That's real usable extra resolution in the 1080p domain.

I used a 50mm Sony F3 kit prime lens at about 5.6 for this... I would expect even more pop with better glass and a fancy down-sample method like the latest ARC ARRI software.

Again, it's a project by project basis and I can't possibly tell you what's right for your project. But ARRIRAW makes a difference for future-proofing AND squeezing that extra bit out of the 1080p final raster. If recording ARRIRAW is difficult and expensive, it's only going to be done on a few high-end projects. If it's no more complicated than slapping a nanoFlash on the back of an EX1, then that sort of re-sets the expectations and demands re-examination of where and when it can be used.

Even if future-proofing isn't on the table, certainly a little resolution wiggle room for upscaling the image and fixing framing is nice, or simply knowing you're giving the VFX guys all you can for that green-screen shoot is enough to keep ARRIRAW in mind...

Regards,

Jim Arthurs

[David Heath]

Originally Posted by Jim Arthurs

Yes, I was scaling Prores to match the frame raster of the native ARRIRAW. And rightly so... how else can we demonstrate the full extent what is lost?

I don't disagree with that, but do believe the point is an important one. There will ALWAYS be a weakest link - and yes, here we see that the camera front end is capable of more than ProRes and 1080p is capable of handling.

That doesn't surprise me - the Alexa is designed first and foremost as a 1080p camera, so I'd expect the front end to be capable of somewhat more than the system. That's sensible design.

But to make use of the extra you demonstrate, we're talking about a recording etc system such as 4k - and then I can see comments like "the camera doesn't nearly do justice to 4k, does it!?" There'll nearly always be a weak link!

Easy fix... Here is a layered Photoshop file so you can pixel-peep the differences. This file is untouched 1080p Prores on the top layer, and the bottom is ARRIRAW downsampled to match the 1080p raster size. No fancy down-sample methods were used beyond simply using the "better for downsampling" option in Photoshop.

What I see is a significant "pop" in overall sharpness in the ARRIRAW, .........

As I'm sure I don't need to tell you, there are two ways an image can "look" sharper - better real resolution and/or using detail enhancement. The big question is whether what you very well demonstrate is due to the former or latter. (Or a bit of both! :-) )

I SUSPECT it's largely down to the latter - and as evidence would say that toggelling to ARRIRAW has a marked effect on the appearance of the lower frequencies as well as the higher. In other words, it's possible that by increasing the in-camera detail settings you may find a value where the differences are far smaller.

This really comes back to my original point. With such as the ProRes option you have to largely "burn-in" camera settings - and that ranges from gain, to colour balance, to dynamic range handling...... and includes detail enhancement setting. Use RAW and all those may be set and adjusted after shooting - including detail/aperture levels. The real benefit of RAW (IMO!) is flexibility - certainly if the final product is destined to be 1080, period.

If recording ARRIRAW is difficult and expensive, it's only going to be done on a few high-end projects. If it's no more complicated than slapping a nanoFlash on the back of an EX1, then that sort of re-sets the expectations and demands re-examination of where and when it can be used.

I don't disagree with that at all - please don't think I'm knocking the idea of RAW recording, quite the opposite. It's just that the original core question posed by the thread was "we are looking for reasons why a production would, or would not choose to shoot in RAW".

My feeling is that the PRIMARY reason why a production would choose to use it would be flexibility. The PRIMARY reasons why they would choose not to would be more work (hence more expense/delay) in post, and a more complicated/expensive recording system. I fully agree that what Dan is proposing may largely negate the second of those disadvantages.

Both Sony and Canon have now brought out large format single sensor cameras with 8 megapixel chips, 3840x2160, with easy to derive 1080p as a here and now, but with 4k RAW as an obvious future. In those cases, RAW offers far greater potential than wih sensors primarily designed to get 1080 by conventional deBayering. (Such as the 2880x1620 of the Alexa.)

(Jim - any chance of reposting the "bunny" test in the same way? Original ProRes, and downscaled RAW?)

[Jim Arthurs]

Originally Posted by David Heath

But to make use of the extra you demonstrate, we're talking about a recording etc system such as 4k - and then I can see comments like "the camera doesn't nearly do justice to 4k, does it!?" There'll nearly always be a weak link!

There's plenty argument that a 4K sensor such as the original RED sports doesn't saturate a projector or broadcast 4K raster either, you need something significantly greater to account for de-bayer issues, but to not record that significant extra detail the Alexa offers you just because it isn't full 4K doesn't make sense in the way that the people who carp about the RED not technically filling that full 4K raster bothers me, it's just a difference in degree.

For the markets the Alexa camera is aimed at; TV series and feature work, every little bit will help. We wouldn't be watching remastered Bluray Star Trek, Twilight Zone and others if those shows weren't "future proofed", if entirely by accident by shooting film. Now with current productions, we have to work at it to ensure that happens and make these options know to those in charge.

Your "weak link" point is excellent, so let's apply it to standard 1080p production. What's the "weak link" in the HD production chain? Any camera/codec that JUST records exactly 1080p. Because it's probably not really giving you the "every pixel different" situation you'd hope for. If it's a 3 chip camera, it's going to alias at frequencies you could sub-sample out with higher res originals. And of course the extra resolution "wiggle room" and value for VFX that I keep bringing up because it's a real benefit.

I agree completely with you regarding the flexibility of RAW being a key asset, but I assert the extra resolution is equally important for 1080p production as well. We're not far off in opinions, I just put a lot of stock in resolution and detail and now that I know what's missing when recording on-camera... well, it's grating. I have an exact metric of the difference, thanks to the test, and know when it would be handy to have on hand, and what to expect.

Originally Posted by David Heath

As I'm sure I don't need to tell you, there are two ways an image can "look" sharper - better real resolution and/or using detail enhancement. The big question is whether what you very well demonstrate is due to the former or latter. (Or a bit of both! :-) )

Back before the RED hit the streets, I was front and center on the Red forums proclaiming that the 4K Red would be a mediocre 4K camera, but an excellent 1080p camera due to the down sampling effect. Down sampling an image that doesn't have edge enhancement produces something which I referred to at the time as "naturally sharp images". What I have given you is exactly that, with full awareness of what I was doing, a "naturally sharp" image with no edge enhancement. It's higher resolving even in the 1080p raster than the Prores because the camera doesn't have the time or the hardware "smarts" to de-bayer and downsample as nicely and with as much computational horsepower as can be done in post.

Speaking of edge enhancements, for fun, throw the "sharp" filter in Photoshop onto each layer and compare. There's stuff that will be accented and brought forward in the ARRIRAW frame that simply isn't there to be enhanced in the Prores image.

The great irony is that it will be easier for independents to take advantage of the benefits of ARRIRAW now that both the recording (thanks to the Gemini) and the post (thanks to Premiere, After Effects, etc.) has been simplified and made vastly less expensive. The high end shows and series have existing workflows they will take convincing to change up or modify.

It really is stupid easy to use ARRIRAW now. One man band stuff... record, transcode and edit.

Do I need it for everything? No, but who says you have to do everything in ARRIRAW on the same show? Use it for the tricky stuff, use it for when you see your crane operator giving you a so-so shot knowing you can de-shake in post and still wind up with a full raster of detail. Use it for the wide shots and the VFX plates.

Originally Posted by David Heath

(Jim - any chance of reposting the "bunny" test in the same way? Original ProRes, and downscaled RAW?)

Sure, I'll process some stuff up when I can get at it on Monday... glad to help.

Regards,

Jim Arthurs

[Sareesh Sudhakaran]

Fine art printing outclasses IMAX anyday, let alone 4K cinema.

What I've learnt from that process is shooting in RAW allows the option of capture sharpening (or deconvolution sharpening) in the RAW program. The algorithm (good ones like Lightroom 4, Capture One, etc) is fine tuned to understand each camera's bayer process (even though it has not been debayered yet) to find the best settings for such sharpening.

Fine art photographers sharpen a second time depending on the image.

And finally, when outputting, depending on the size of the print, paper quality, web, etc,. they apply an output sharpening as well. Third time, here.

Arriraw, being RAW in the same tradition as digital still cameras, allows this excellent opportunity in video. But:

My question is, there are no algorithms produced specifically for Arriraw to replicate this advantage. At most, I see software "supporting arriraw natively" - whatever that means.

I would pose the same question of Redcode Raw, for which Red has released specific software based on their sensor and codec. Is there something like that exclusively for Arriraw? As far as I know, the originally authorized arri recorders just recorded .ari files, without any processing.

The scary part is, if there is none - then the sharpening applied by various plug-ins and algorithms on arriraw are the ones used for full raster RGB images - which might improve the image, no doubt - but isn't the best possible solution for RAW.

Just my random thoughts on the subject.

[Dan Keaton]

Dear Sareesh,

I enjoy reading your posts.

I do not feel fully qualified to answer all of your questions or comments. So I will leave that to others more qualified.

I can state that there are three main ways to start using ARRIRAW.

1. The ARRIRAW Converter, available for free from ARRI. This is a Mac based program.

ARRI Group: ARRIRAW CONVERTER

The ARRIRAW Converter was recently upgraded.

2. ARRI offers a SDK, a Software Development Kit, which is used by many professional software companies.

This was also recently upgraded. One significant upgrade is the ability to use the Graphical Processing Unit to speed the deBayer process.

3. Software companies can write their own code and algorithms.


Side Note:

Before I started researching RAW, I naively thought that there was one, proper and correct way to deBayer an image.

I have leaved that this is not the case.

One can do a quick deBayer for confidence monitoring.

One can do a better deBayer, wth more time and computational horsepower. And this may involve multiple passes.

And one can use state-of-the-art deBayering software.

This will obviously provide the best deBayer.

But, an important point is the "state-of-the-art" advances over time.
So if one archives the original Raw files, one can go back later and re-deBayer the original files and obtain better images.

ARRI's new SDK should, in my opinon will have a very positive impact.

1. New, more sophisticated deBayering algorithms are used.

2. The GPU, (Graphics Card) on your computer can be utilized to provide vastly increased computational horsepower. Thus, a very compute-intensive algorithm (computer code) can be used to provide a better image in a shorter period of time.

[Sareesh Sudhakaran]

Hi Dan, thank you for the kind words - I'm usually guilty of having strong opinions and I'm grateful to anyone who puts up with it!

I can guess how tough it must have been for your engineers to make sense of the voodoo data stream shoved through a dual HD-SDI cable. My line of thought isn't a reflection on the performance of the Gemini 444.

Here's an example of something that probably supports my point of view:

The Arri Converter/SDK has two debayer algorithms - ADA-3 HW (Hardware) and SW (Software) - none of which provide any degree of control except -
Sharpness from 0 to 300 (Default 100). This does not correlate to any traditional and professionally used sharpening algorithms.
The maximum color bit depth rendered to DPX/TIFF is 10/16-bit - no 32-bit float option.
It has excellent support for color spaces, but very few 'dials' for color control. If I shoot green chroma and need to tweak the green color, I'll have to use a 3rd party grading application. So why bother with the converter?

The Arri converter basically mirrors the in-built camera controls. It's like Henry Ford's famous "any color as long as it is black" policy on the Model-T.

Lazy Arri? I don't know and I don't want to judge. How can I? I love the Alexa as it is, warts and all.

But real-time raw processing seems like the next big challenge in video! In the coming months, it will be interesting to see how other manufacturers deal with these issues.

[David Heath]

Originally Posted by Jim Arthurs

Down sampling an image that doesn't have edge enhancement produces something which I referred to at the time as "naturally sharp images". What I have given you is exactly that, with full awareness of what I was doing, a "naturally sharp" image with no edge enhancement. It's higher resolving even in the 1080p raster than the Prores because the camera doesn't have the time or the hardware "smarts" to de-bayer and downsample as nicely and with as much computational horsepower as can be done in post.

"It's higher resolving even in the 1080p raster than the Prores because the camera doesn't have the time or the hardware "smarts" to de-bayer and downsample as nicely ..." - maybe.... but it could be because the in-camera detail setting is set low. Maybe to a negative value to deliberately knock detail off for a "look"? That's what I meant by "it's possible that by increasing the in-camera detail settings you may find a value where the differences are far smaller".

OK, that's maybe a bit academic. The real point is that with RAW the detail setting is changeable in post, with ProRes it's burnt in, so no question - the RAW gives you flexibility.

[John Brawley]

Originally Posted by Sareesh Sudhakaran

What I've learnt from that process is shooting in RAW allows the option of capture sharpening (or deconvolution sharpening) in the RAW program. The algorithm (good ones like Lightroom 4, Capture One, etc) is fine tuned to understand each camera's bayer process (even though it has not been debayered yet) to find the best settings for such sharpening.
.

You might want to try Davinci Resolve. You can download the free version and it will work with ARRI RAW as well as RED RAW and DNG natively.

The free version is exactly the same as the full version, except it has no noise reduction, 3D support and is limited to a single GPU and 1920 resolution (on export).

It has a very good Debayer engine and Resolve is probably to "go to" motion RAW grading tool at Hollywood / Studio level productions.

It's kind of amazing you can get the same software for free.

jb

[Sareesh Sudhakaran]

I agree, John. 100%. It is really cool of them to offer it for free.

I'd love to try Arriraw on it. but everyone here shoots Prores. Maybe I'll get around to testing it some day. Thanks for the tip!