Shooting in log?

[GlennChan]

Stu,
Whaddabout this:

A- When doing the test, shoot at a number of different exposures / bracket your exposures.
B- If the camera had camera settings, you might play around with them. You might also experiment with optical filters. But for simplicity's sake, let's forget about this. (And camera settings aren't important for Red since it shoots RAW.)
C- Bring them into your grading suite, grade the images.
Redcine settings would also be part of this. For simplicity's sake, it might be prudent not to do Graeme's highlight recovery (since it won't always work; but if it does work, then it's a bonus).

When looking at the final result, you'd be able to figure out your exposure "sweet spot". In production, you might under/overexpose by a stop due to production pressures (or situations which you didn't test for) so you want to figure out if exposing with a safety margin is necessary.
But anyways, by bracketing your exposures, you'd figure out your ideal exposure.

Once you've figured that out, you would try to relate it to your monitoring tools. You want to be able to use your monitoring tools to get the ideal exposure. It might be that 3-d LUTs are ultimately the best monitoring tool? (It's what you'd be using the DI suite anyways right?)
David Mullen's suggestion might be very helpful in the meantime.

 

[Stu Maschwitz]

We haven't heard from Graeme yet on this thread, but I have a question/request that I think might be on his turf.

In RED Alert (and I think RED Cine as well, although I have reverted to RED Alert for stability reasons), when I select PD Log 685, those blown-out windows are clipped at 685. But when I drop my Exposure down a bit, a little more highlight information creeps in.

I know Graeme has often recommended dropping exposure to extract all the available values from the raw data, but in this case I would not expect to need to do that, since there's plenty of room between 685 and 1023 for those values. In other words, I wish RED Alert could map those values into the registers above 685 along the log curve, without requiring me to adjust exposure.

Does that make sense?

-Stu

 

[David Mullen ASC]

I'm not clear on the exact procedure, but one test would be to ignore the 720P Rec 709 monitoring output and just shoot a series of greyscales at different ASA ratings (not bothering to change the rating in the camera since that's just metadata), then look at the conversion from RAW to Log RGB on a waveform and pick the ASA rating that you like best for the range being captured, then go back and actually set the camera to that ASA rating and see how the Rec 709 preview looks on a waveform compared to how the RAW-to-RGB version looked on a waveform in post.

 

[Mike Prevette]

Stu,

Fantastic read over at your blog on this. I'm afraid after reading it, and playing with the Red as little as I have that the mid tones and shadow data just couldn't survive a 2.580 stop under exposure.

After thinking about it from a non scientific, shooter standpoint I seem to be of the mind that in situations I want shadow detail, I would overexpose, and in ones where I want highlight data I would underexpose. For example this week I shot tests ona beach in midday sun, and in an alley way at dusk. Both of those scenes are very different in tone (high key vs low key) and they would have benefited from me having more highlight detail in the sunny one shadows be damned, and more shadow detail in the dark one highlights be damned.

 

[Graeme Nattress]

There are some DI apps that igorne brighter Cineon values, or so I'm told, so Log 685 plays it safe. If you want them to occupy the higher values, the 985 version is for that.

In camera, there's some 0.4stops exposure headroom, and hence the need, unless you've altered in camera settings, to set the exposure a little lower to get the full range out.

Graeme

 

[Dj Joofa]

But the Rec 709 output is controlled by what ASA rating you set the camera to, even though that's only metadata. So if you set the camera to 1000 ASA, for example, you'd still find on Rec 709 monitoring signal the same clip point, but on the RAW files, you would have underexposed the footage and thus gained, let's say, a stop and a half or two stops more information in the highlights, at the expense of information in the shadows. Not to mention more noise overall once you correct to normal brightness.

David, interesting. I have a question. I can understand what is happening on the negative when we say in film terms say 1000 ASA and under/over exposing on the negative. But according to your comments and information elsewhere on this forum, the Red camera is only recording corresponding information in the metadata, and perhaps not making use of it until the raw signal is out of camera into a processing tool such as RedAlert/RedCine. Then it means that all information is there in the raw file, and when you say something like "at the expense of information in the shadows", does that mean that RedAlert/RedCine is delibrating getting rid of information that is there in the raw file just to simulate real world behavior with a film negative in a film camera.

Similarly, I have little confusion when you say "on the RAW files, you would have underexposed the footage", because if the raw file has no processing applied to deliberately underexpose raw signal (which a film camera might do optically on a real film negative), and only in post-processing through RedAlert/RedCine, you are simulating under/over exposure, then it would give me an indication, that again, RedAlert/RedCine is either creating information which is not there in the raw file, and/or destroying (i.e., not making use of) some information which is in the raw file.

Any clarification help will be much appreciated. Thanks.

 

[Stu Maschwitz]

There are some DI apps that igorne brighter Cineon values, or so I'm told, so Log 685 plays it safe. If you want them to occupy the higher values, the 985 version is for that.

I think you've been lied to—I find that very hard to believe. No DI system I've ever seen ignores any Cineon values within the 0-1023 range.

Besides, if it did ignore them the results wold be no different than if you clipped them, so why clip?

In camera, there's some 0.4stops exposure headroom, and hence the need, unless you've altered in camera settings, to set the exposure a little lower to get the full range out.

I get it, but why clip internally? Why not just pass the full range through regardless of which white point I select for log? Clipping is something you should really avoid, for so many reasons.

I guess in the meantime I'll just create a setting that uses 985 and has a healthy enough exposure drop to include any headroom. But that feels like quite an opaque hack. It would be very nice to have one setting that mapped what appeared as white to a consistent value and passed your lovely headroom along as values greater than "white."

-Stu

 

[Jon Corcuera]

Since we are talking about expoition an latitude I find interesting the idea of noise, is ok to take the decition of what noise level is good or which is not but, just by eye. But i have been a while wondering if there is a way to mesure the noise(grain like) that apears, is there a sofware that does it?
Jon Corcuera

 

[David Mullen ASC]

David, interesting. I have a question. I can understand what is happening on the negative when we say in film terms say 1000 ASA and under/over exposing on the negative. But according to your comments and information elsewhere on this forum, the Red camera is only recording corresponding information in the metadata, and perhaps not making use of it until the raw signal is out of camera into a processing tool such as RedAlert/RedCine. Then it means that all information is there in the raw file, and when you say something like "at the expense of information in the shadows", does that mean that RedAlert/RedCine is delibrating getting rid of information that is there in the raw file just to simulate real world behavior with a film negative in a film camera.

The camera has an "x" number of stops of dynamic range captured in RAW -- what I meant was that when you treat the camera as if it were faster, more sensitive, like at 1000 ASA, you are basically favoring brighter areas at the expense of darker areas. The number of stops of information recorded is the same, it's just a matter of which end of the scene's brightness range you want to favor. Of course, though, you have to later color-correct this footage to the final luminance values you want, which is the problem, because noise becomes a factor if you start having to brighten the image. But you biased the exposure towards the shadow range of information, then the problem is clipping in the highlights.

So essentially it's the eternal conflict between noise vs. clipping, until the dynamic range capability itself is expanded. Of course, not every scene is pushing the limits of dynamic range.

No, there is no "getting rid of information" in post unless you want to get rid of it. Besides, since film negative has more dynamic range than what the RED can capture, you wouldn't be throwing away information necessarily to match a film look.

However, in my LART test, I did find that RedAlert was getting rid of highlight information when you shot at a higher ASA rating and then opened up a file that higher ASA ratings, because it was doing a simple "gain" boost. It wasn't applying a gamma curve to keep the dynamic range captured in RAW. So it was better to open up the file at a lower ASA value, even though it looked dark, and then color-correct it "up" to normal by changing the gamma curves. I don't know if REDCINE handles the ASA metadata better in that regards.

 

[I Bloom]

Stu,

I really enjoyed your Prolost post. I'm glad you've started this conversation and that you've done so with such acumen.

One of the things that has bothered me about Red is that the waveforms and histograms shift when you change the metadata. This doesn't make terrible sense to me. Especially in the case of the incredibly useful false color meter. It seems to me that the meter should always show you exactly how much detail is being recorded in the highlights and exactly how much the shadows are in the mud and those two things should never change. Regardless of ASA. The only thing that might change then is the position of the 18% Gray (Green) level and the position of the Skintone (Pink) level.

We should see the same markers, green and pink in the histogram and waveform, and these should be what shift, not the graph itself:

IBloom

 

[Stu Maschwitz]

Hi ibloom,

I have to say, that is the conclusion I'm coming to as well. We need a way to preview everything the sensor is recording, with meaningful clipping warnings and exposure targets.

-Stu

 

[David Mullen ASC]

It's because you're measuring the 8-bit 4:2:2 720P Rec 709 picture signal generated for monitoring purposes. I don't think it's possible to read the luminence values of a live RAW image pre-conversion, but I could be wrong. I guess one solution would be to have a separate signal for displaying on a waveform that is different than what goes through the gain adjustments and whatnot for monitoring, or have one minimally processed 720P signal but be able to measure that signal before it gets processed further for looking correct on a monitor.

As others have suggested, selecting an ASA setting that is as close to the "true" sensitivity of the sensor should get the waveform and false color readings closer to being representative of the RAW image, but this is a problem for low-light scenes where you have selected a higher ASA rating and want the monitor output to look correct in brightness. You'd have to switch back and forth between the ASA rating desired for the scene and, when you want to double-check for clipping and whatnot, the ASA rating that gives you a better idea of what's being recorded in RAW.

 

[Dj Joofa]

The camera has an "x" number of stops of dynamic range captured in RAW ...

Thank your very much, David. I really appreciate that. I have read numerous of your posts and I must say that your command of the English language is very good. The way you write and explain things using the right and minimum of words required is certainly very helpful and didactic. Does it have anything to do with your BA in English?

 

[Arnaud Paris]

I'm jumping in this very important conversation because what you guys are pointing out here is the key to this camera I believe.

We've been using camera 88 since November and we've shot several projects for filmout going through our Scratch, so we faced the dilemna of noise vs clippings a number of times.

The solution to this preview problem will come soon when build 16 is released with LUT capabilities. You then will be able to load a LUT file in the camera that will affect the monitor preview of the camera affecting the histograms as well.

So by making a custom LUT that reproduces the log curve of film you could actually light your scene with a somewhat accurate histogram on the camera for filmout.

We don't expect Red to give us a video tap from the actual raw sensor image (which monitor would be able to handle that anyway). What the viper is doing with this raw preview mode is actually the same thing I described just before: putting a LUT in the video feed to reproduce the log curve characteristics.

So I guess until build 16 comes out, or if build 15 has a "raw" preview simulation available, we'll just have the DP test before the shoot how many stops he has left when the 709 monitor signal clips. By knowing this he can light his scene using the histogram and his lightmeter. We've done this on the last 4 projects and it works just fine that way.

Just so you know, most of the DPs have decided to rate the camera between 160 and 250 in regards to the acceptable noise floor and they measured that when the REC709 signal clips on the monitor (reaches 108IRE if you want) they have about 1.5 stops left in the brights.

 

[Gunleik Groven]

thanks Arno.

That was very usefull.

gunleik

 

[Stu Maschwitz]

It's because you're measuring the 8-bit 4:2:2 720P Rec 709 picture signal generated for monitoring purposes. I don't think it's possible to read the luminence values of a live RAW image pre-conversion, but I could be wrong. I guess one solution would be to have a separate signal for displaying on a waveform that is different than what goes through the gain adjustments and whatnot for monitoring, or have one minimally processed 720P signal but be able to measure that signal before it gets processed further for looking correct on a monitor.

As others have suggested, selecting an ASA setting that is as close to the "true" sensitivity of the sensor should get the waveform and false color readings closer to being representative of the RAW image, but this is a problem for low-light scenes where you have selected a higher ASA rating and want the monitor output to look correct in brightness. You'd have to switch back and forth between the ASA rating desired for the scene and, when you want to double-check for clipping and whatnot, the ASA rating that gives you a better idea of what's being recorded in RAW.

Unless of course there was a "true zebra" that showed you chip clipping rather than display clipping.

-Stu

 

[David Mullen ASC]

Unless of course there was a "true zebra" that showed you chip clipping rather than display clipping.

-Stu

Well, that's sort of my question -- is it even possible to read / measure that (clip point) in a RAW file, live, without converting it to RGB?

BTW, technically the Viper isn't a "RAW" camera because it is not a single-sensor Bayer-filtered camera -- what it produces is a minimally processed 10-bit 4:4:4 signal with a LOG emulation from the 3-CCD output.

The notion of rating the camera at 160 to 250 ASA for the noise level and having 1.5 stops of overexposure latitude is what I was referring to in regards to finding a balance between noise vs. clipping that you can live with. To some degree, noise can be adjusted/fixed in post, but nothing can save a clipped highlight. I would suspect that the 320 ASA rating some people are using is probably a better compromise than 160 ASA would be.

As regards to my writing skills, I hope those years of writing English papers at UCLA had some practical value... :wink:

 

[Dj Joofa]

Stu,

IBloom

IBloom, since your graph was not labeled, I could not figure out where were you putting that all so elusive "18% gray.". I think it will be easier with an example. Suppose that the minimum black in the scene is 1% bright and the brightest object is 1000% bright object, then, are you putting 18% gray at:

(1) 18 times higher than minimum black, and hence, at 18. Or,
(2) 18% of the full tonal range of ~1000. I.e., 1000 * 18% = 180. Or,
(3) Elsewhere.

In addition, IMHO by fixing skin color/tones using a histogram, you would be doing in effect (and not in theory) the same thing which NTSC committee did more than 50 years back when they fixed I axis of the I-Q axes, and hence, may not work in all situations.

 

[Stu Maschwitz]

Just so you know, most of the DPs have decided to rate the camera between 160 and 250 in regards to the acceptable noise floor and they measured that when the REC709 signal clips on the monitor (reaches 108IRE if you want) they have about 1.5 stops left in the brights.

With Paul's footage, if you set RED Alert to what we metered at, ISO 320, there is indeed a touch more detail in the windows that can be extracted via lowering ISO or Exposure. But not 1.5 stops—more like about 0.4 stops, which is consistent with what Graeme has said here before (about how much to lower Exposure in RED Alert to ensure that you grab all the love).

See this screen session:
http://rebelsguide.com/dl/windowExpo.mov

For this particular footage, if you rate at 250 you get what looks like all the chip had to offer. If you rate at 160 you don't get anything more. Another way or performing the same test is to reset to 320 and drop Exposure—drop 0.2 stops and there's still more to go, but drop 0.4 stops and once again you've revealed all there is.

So for this camera on this day with the winds blowing as they were, it would appear that you have about something less than 0.4 stops of headroom (in this case defined as useful raw pixel values brighter than what you are previewing via Rec709 video out of the camera) when rating at ISO 320. Had we rated at ISO 160, it would seem that we would have no headroom at all—we would have been previewing the entire usable range of values captured by the chip.

-Stu

 

[Dj Joofa]

I don't think it's possible to read the luminence values of a live RAW image pre-conversion

Not the ordinary users. However, I am sure Red can have that data if they measure and calibrate their sensor internally.

Well, that's sort of my question -- is it even possible to read / measure that (clip point) in a RAW file, live, without converting it to RGB?

I think what is clipped is in reference to the white point of the space you are going to view the raw data in. Hence, unless you fix a space in which you are going to view raw values after raw->RGB conversion (say Rec 709), it will be difficult to access what will clip.