Understanding ISO with the RED ONE

[Dj Joofa]

allright i'm confused already. i learned the zone system through B&W still photography and from what i know 18% grey is 18% grey, that is to say if you want to realistically render a scene, a grey card is in zone V by definition, regardless of ISO.

Yes, you are right. All this confusion is arising because people are freely mixing different terminologies. If people just think of ISO setting on a digital camera (Red inclusive) as a gain knob then the confusion may go away. Please forget about the ISO issues of the film world. Just think that when you are increasing ISO on your Red you are asking for positive gain be applied and when you are lowering ISO you are asking for negative gain. Think of it like that gain knob setting on a typical camcorder.

 

[Andrew M.]

Hi Andrew,

we did some research today.

According to our own search results and the current entries on

http://en.wikipedia.org/wiki/H264

we came to the conclusion that there is currently NO software that supports 11-14 bit high profile 4:4:4 H.264 encoding at all. See the lower part table in the wikipedia.

Even 10 bit seems to be almost out of reach.

I really wonder which software you are using there, that gives you these options...

The X.264 codec allows us to select many options that would yield towards the 14 bit capable profile, but the important setting is kept in baseline, so ending up in 8 bit output too.

We found two major issues with the H.264 encoding as of now:

1. The slight gamma shift, which is driving us nuts over here, as the very same file is having at least 3 different looks in playback on the very same incarnation of Quicktime, depending on how you play it etc.

2. The artefacts on lower quality settings. The 75% setting we used for COLD STORAGE teaser was almost the lowest end we can go, without further sacrificing the look with artefacts.

We did today encode a music video at 75% with H.264 and it was really bad experience, the background was a crappy ocean of banding. All the subtle gradients from the shooting where dying there. We'll try again tomorrow with higher quality settings to see how or if we can safe it.

What is your experience with brightness / gamma shifts in H.264 or QT in general?

Axel

Maybe we should take the codec discussion to this thread and leave ISO.
I will answer your questions there.


http://www.reduser.net/forum/showthread.php?t=13130

Andrew

 

[Blair S. Paulsen]

ISO is just a state of mind

ISO is just a state of mind

I agree. As fascinating as codec discussions inevitably are, this thread was about how the ISO setting on the RedOne may not work the way one expects. Furthermore, there have been some interesting notes on different ways people have chosen to approach the basic issue of "proper" exposure.

Let's see if we can identify some basic concepts that will help improve the quality of everyone's work.

Mac started us off with some test images and asserted that by ETTL one could get a more filmlike shoulder with the caveat that ETTL brings the noise floor up into higher zones. I think the consensus was that for situations with 5 stops or less of DR this could work well.

ETTR (exposing to the right) was touted as the best way to get the most DR out of the RedOne and to minimize noise.

The most effective way to manage the on camera ISO setting for monitoring and for the "first look" in post seems to be in the eye of the beholder and often beholden to the post workflow.

So if we stipulate that a big portion of the ISO setting discussion is more about signal to noise ratios and one's stance on clipping artifacts we find ourselves in territory familiar to all photographers. Make an informed choice between ETTL and ETTR based on the situation/project and shoot on.

The one remaining issue in my mind is non-linear nature of the noise. Assuming you plan to crush Zone 1 the real battle is in Zones 2 & 3. I have been amazed at how much more noise there is one image compared to another one that wasn't all the different based on normal measuring procedures (light meter, on camera histo, etc).

There have been some valuable threads on the quantity of noise in the blue channel. If you want to keep noise down and can pick your lighting color temp I feel safe recommending a higher Kelvin option; 5000 plus. This is also helpful is you hate clipping in digital because the lower noise floor with the blue channel well fed will allow you to shift your histo more left.

In a perfect situation I would light daylight balanced, use softer sources, manage specular highlights, keep everything in a 7 stop range and still ETTR just a little. Once in Scratch/RA/RC I would then carefully bend the curve to get the zonal distribution that suits my intention for the shot.

Ultimately my biggest focus on set is getting the the right amount and quality of light to fall where I want it. Hard light, soft light, toned light, haze, soft and hard cuts, etc are still how we paint the scene. Same as it ever was.

 

[Darren Orange]

Maybe I am off here, but what stops you from really having the best of all worlds? Could you not take that clip and with Scratch bring up the ISO of one part of the clip and maintain the levels on other parts to prevent unwanted noise? Perfect use for scaffolds I would think.

 

[Darren Orange]

Here is a link to a mock up idea, it shows kind of what I am think could be done in real time with something like Scratch. You can download it here, right click save or stream it.

Example Mixing ISO

 

[Jason Sinclair]

I'm thinking the same way as you. I am hoping build 16 and a software change to redcine will achieve this. It's the only way i can see of getting as much out of the sensor as possible for as little trouble.

 

[Jason Sinclair]

 

[RedRay]

wow reading through this thread is really like taking a cinematography workshop allover for me Great opinions expressed here! I'm really learning a lot. For almost 25 years I've been shooting on film format, from super-8mm to 16mm to 35mm. I've only somewhat slowly converted to digital cinematography in recent years, having used the Sony F900, Panasonic varicam, and my own Sony Z1 hdv. At the end of the day, I hope it comes to shooting with the Red as close to our old habits, although it is obvious here that that is not possible, not yet. For years I shot film without, or with a very bad video assist monitor, and it did not matter, because i learned to see how my chosen film stock sees. I knew when to kick in fills, add reflectors, adjust backlights, use filters, or choose the time of day to film, etc. because i could see the image in my mind. I can not claim to know exactly how it is registering on the film emulsion, but you knew that you had an image you could push and pull when you got to the lab color timing, or the telecine suite. Of course, you never stopped re-learning as Kodak and Fuji constantly come out with new improved film platforms. I hope to be able to do that with the Red, to learn how it sees and discover your comfort zone, knowing that you got the shot you want, even if you only had that tiny attachment lcd to look at.

 

[Chris Kenny]

Maybe I am off here, but what stops you from really having the best of all worlds? Could you not take that clip and with Scratch bring up the ISO of one part of the clip and maintain the levels on other parts to prevent unwanted noise? Perfect use for scaffolds I would think.

Yeah, once you get into post, the ISO setting isn't that useful. Why brighten or darken the entire image, when you can have much more fine-grained control, even with with just the curves in Redcine?

What is significant, though (and what most of this thread has been about), is what assumptions people should make about the camera's ISO when determining exposure on-set. This will probably become much less of a guessing game if build 16 can display histograms and false color exposure based on the raw sensor data.

 

[Andrew M.]

I find these color temperature and ISO bit confusing and kind of technology of the past.
Though we grew up with it so we need to relate to something at the beginning.
Good 3 color histogram and these traffic lights from firmware ver 15 is the future.
We need new tools for determining exposure. Something specially made for RAW digital.

Anyone has good idea how to revolutionize the Sekonic?

Good articles.
http://daystarvisions.com/Docs/Tuts/DCExp/pg1.html
http://www.sekonic.com/news/reviews2.asp

 

[Daniel García]

Here is a link to a mock up idea, it shows kind of what I am think could be done in real time with something like Scratch. You can download it here, right click save or stream it.

Example Mixing ISO

I think you are making a big mistake here, and please correct me if i am wrong.

What you are compositing here are not two different "ISO" interpretations of the same shot. That would be feasible, although not anything philosophically different over using a curve or more complex processing on the full sensor data.

What you have done in that shot is compositing two different shots, made at different exposures to account for the fact that they were to be interpreted at different "ISO's" in post.
Combining different exposures to achieve a wider latitude is the basis of HDR photography, but not applicable to moving images. At least with current technologies and methodologies.

 

[Jason Sinclair]

No, These shots we have posted are using the iso shots. There was no change in exposure that i am aware of. (I just used the shots from the wav files so better results would come from raw)
In the case of the one i used there was three iso ranges which in affect pulls the two ends of the spectrum out by a) getting more details out of the highlights in the high value iso and then b) getting cleaner shadows out of the lower end iso and then adding these values (minus the unwanted data) to the mid range iso thereby squeezing the range into a viewable format. Try doing that in curves... Philosophically you can but curves doesn't have the correct way of doing it because it's more complex as it needs layering as well as adjusting contrast and light levels.
In essence you are finding hidden information on the raw data on both sides of the range which at the moment we can't do with our eyes because theres no 4 k monitors available. It's also a matter of contrast and how human eyes perceive detail. The sensor cant natively capture and perceive contrast as the eyes can. Like the human eye system, perception has to be processed in a brain. It has the advantage of taking that information back to the eyes and readjusting them for the different scenario. Here we are just pecieving the raw data and i would prefer to do this in red cine because the more processing power you have the more you can do right? However, yes pre raw data would have a lot more info in it. If they find a way of doing this then... It would be interesting to test a 4k uncompressed frame and see how much detail you would get out of it. I'd say you would actually get a lot. Their genius never fails to surprise me and some work could be done here i'm sure but equally using that raw footage and then going at it with 8 cores to work those numbers in their complex algorithms to pull the detail together in the best perceivable image would be a dream come true. In essence, it's a codec issue but i gather something more complex. I'm out of my league to discuss this.

Just because you can't see the dr doesn't mean it isn't there and so we need a process which brings it down to a viewable range. With film it is scanned into digital and so even though downrezed, it retains the most of it's percievable dynamic range visually or more to the point comparatively, but it is essentially the same thing. Same when you downrez that to DVD, the codec compresses the range into a viewable format. I can watch a good movie on vhs and still enjoy it but i can't watch a movie shot on vhs and enjoy it.
So you are right on most accounts but don't think reactor88 is making a mistake because he's pulling a better image as that is the name of the game and philosophical construction comes second.

 

[Dj Joofa]

I think DaNni does have a point here as Red is not doing (almost) simultaneous multiple exposure to get (a form of) HDR. Therefore, amalgamation of images acquired at different exposures at different times is not what you would typically get in a continuous shoot.

 

[Darren Orange]

The exposure was not different aka iris or f stop. It is the same shot shown with 3 different interpreted ISO. The ISO is meta data and does not change the way it was reordered. Recode RAW records the RAW data. You can interpret the RAW at different "ISO" in post. So you can keep your DR high and noise low IF you expose the camera based upon a rating of 320ISO. You should be able to bring back the high lights and keep your shadows and blacks noise free. It would be great if we could get a Redcode RAW version of this file to play with. Maybe this is the best logic on this topic, You can Raise the ISO beyond 320ISO of blown out parts of the screen to get back more DR without getting any more noise BUT you can not do this with Black to get back more DR without getting Noise with it.

Thats what it looks like but I can't tell without a RedCode Raw File....RedRelay here I come! :w00t:

 

[Dj Joofa]

Maybe this is the best logic on this topic, You can Raise the ISO beyond 320ISO of blown out parts of the screen to get back more DR without getting any more noise BUT you can not do this with Black to get back more DR without getting Noise with it.[/COLOR]

Reactor, this whole thread is a little confusing for me as I am having trouble with statements such as "You can Raise the ISO beyond 320ISO", without additional information. May be my understanding of Red needs correction, but here is how I see it.

When you raised ISO beyond 320, it is not clear to me which one of the following two options were chosen:

(1) No metering/aperture size/exposure time/etc. was performed.
(2) Corresponding metering was performed.

In case of (1) when you just simply raise the ISO without any corresponding metering to go with it, then camera does not care, and passes along signal values to software. Software ISO gain will boost up the signal values and you might:

(a) Loose highlights
(b) Increase noise in darker areas.

In case of (2) where corresponding metering was done to go with an increase in ISO, the following happens as I understand:

(a) The actual recorded signal is reduced as aperture is reduced, or exposure time is reduced, or any combination there of. Since Red is only recording ISO boost in metadata, the camera is not controlling the signal. However, the physical metering applied reduces the signal values. As a result:

(i) Blacks/darker areas are crushed.
(ii) More highlights are pulled in (the courtyard outside that window in the image), which otherwise would have been missed. However, since metering is reducing signal values (and the camera does not care), and so previously uncaptured highlights are pulled into normal range.

(b) The camera passes these values along to software ISO gain. Software ISO gain now actually does the reverse to some extent and:

(i) Raises highlights. Some highlights might get knocked off, but enough might have been pulled in because of the metering and some of that "extra" highlights still remain.
(ii) Increases some noise in the darker areas.

 

[Darren Orange]

Well...

Well...

Okay, I think I can make my thinking more clear. Sorry for the confusion.

Light your Black or Shadows practically and don't let your highlights blowout. They might look blown out at 320ISO but test to see what you can get back at something like 200ISO. I would alternate Between Blacks and Shadows looking right for 320ISO and making sure you keep plenty of highlight detail at 200ISO. That would be my game plan. You should be able to see all your possiable mixed DR via viewing what your recording at ISO 200 to 320. Again the LENS controls of exposure is part of LIGHTING. You have to light to accommodate the lens and the lens can accommodate the lighting. If your Blacks or Shadows are not right for you open up the lens or add more light. Just be careful of what you are getting in the highlight areas. By Blending two different ISO settings on same shot and take you should see an increase of DR by maybe 2-3 stops tops, at least 1 stop.

I am going to do more testing I will post examples and results for everyone to see.

 

[Daniel García]

Please check the facts

Please check the facts

Oh please. What a mess.

Supernovafilms and Reactor88 I have to ask you to check your facts before you post. When someone shows a disagreement with you, revise your logic before you make multiple wrong posts. Please don't be stubborn.

No, These shots we have posted are using the iso shots. There was no change in exposure that i am aware of. (I just used the shots from the wav files so better results would come from raw)

The exposure was not different aka iris or f stop. It is the same shot shown with 3 different interpreted ISO.

And now let's go back to Macgregor's second post (emphasis mine):

Again, by changing the ISO we just move the mid gray accross the zone system of our image, making our 709 image displayed in the LCD appear correctly (but affecting dynamic range). But since we are exposing accordingly to the ISO, the amount of light that will hit the sensor will not be the same.

If you are basing your analysis on the shots from iso.wmv, these are different shots indeed. You can even check by eye that the background movements are not the same.

I understand that this is a complex subject and it's easy to confuse some things if you don't pay a lot of attention. Sometimes you really have to read the same things twice just to be sure of what we're really talking about. So please check your facts before you start pontificating/theorizing and worst of all... confusing the rest of the people even more. Don't add to the noise.


Actually, the idea is not that complicated. Macgregor has tested empirically what Stu had theorized about.

Fact #1: RED has no shoulder/rolloff into the highlights because it's a linear sensor. It clips abruptly. To achieve the same look as film we have to underexpose and then apply a curve that imitates film's behavior in the highlights. Underexposing pushes the signal towards the shadows, where most of the noise lives, and then it'll be lifted by the film curve. So this technique will be limited by how much noise we can accept to lift.

Fact #2: ISO in the Film world equals sensitivity/coarseness of the stock. In the SLR world it's the analog gain applied to the sensor data before it's digitised. In the RED world it's the metadata that's used for the interpretation of the raw sensor data, untouched. This interpretation affects the live monitoring and can be changed to any other interpretation in post.


#1 + #2 = Eureka! We can actually use RED's ISO control for something useful! [1] We can apply the film-look-resulting underexposing strategy on set AND not have to look at a dark image!

That's all there is to it!

Now what Macgregor tests provide us is an empirical data point that gives us an idea of how much we can underexpose and not sacrifice too much image quality. It looks like ISO 500 is an acceptable point, and at ISO800 the road starts to get bumpy... but, by all means, DO YOUR OWN TESTS and determine the amount of noise you're prepared to accept in your circumstances (daylight/tungsten, etc).

I hope that's clear and short enough
A lot of this info is bound to change with the release of build 16, with significant changes to dynamic range, signal processing, live monitoring... we can only wait!



[1]It's just an annotation, it doesn't touch the signal, you're always ISO320. In that sense, IMHO the DP has lost a tool for effectively controlling his exposure, and I think that analog gain gives the DP more flexibility. Graeme has talked about analog vs digital gain in fxguide's red centre podcast #4, and he recognised that analog gain has its strong points. I hope RED changes gears and the Mysterium X gives us that opportunity. But that's just me, and it's a subject for another thread.

 

[Graeme Nattress]

I've tested comparing analogue and digital gain. Common wisdom says analogue gain looks better. I tested it, and on my Canon DLSR it did. But I've shot both my DSLR and Red set to ISO800 and had the RED look better.... (used a light meter on the scene, set both cameras to what the meter told me) My conclusion is that digital gain, if done well, and done with good code, can look good

Graeme

 

[Dj Joofa]

I've tested comparing analogue and digital gain. Common wisdom says analogue gain looks better. I tested it, and on my Canon DLSR it did. But I've shot both my DSLR and Red set to ISO800 and had the RED look better.... (used a light meter on the scene, set both cameras to what the meter told me) My conclusion is that digital gain, if done well, and done with good code, can look good

Graeme, it appears you tested them on different systems. I have done testing analog and digital on the very same systems and analog gain has some very good advantages in hiding posterization, especially in low light scenes. It is amazing that in really low light how much the analog gain is better than digital gain. In normal to high light situations the differences are less pronounced, and in fact in some cases, digital gain gives a very clean image.

I found an adaptive workflow that consists of combo of analog and digital gain that is hard to beat.

 

[Daniel García]

That was great having that fast and direct answer to a little tangential remark. Thanks Graeme.
Talking from a theoretical standpoint... I guess the biggest risk with digital gain is posterization, due to only using a smaller part of the code values. How do you fight against that? Dithering maybe?
This reminds me of your comment in that same podcast about... when you are in ten bit space, even if you don't use a big part of the range, you still have a significantly bigger space than on 8 bit (thousands of values).

So I guess the logic would be... the more you underexpose the less of the 12bit range that you use and the less elbow room you have to push the image around in post. 4 stops under gets you an 8 bit "exposed range", and an iso2560 equivalent.

I had never thought of it this way. Am I wildly wrong?