ISO of Red

[Dylan Macleod CSC]

Interesting- it looks like I need to do some more experimenting with this ...

It was my understanding that the sensor performs the best when set at ISO 320, and that processing the image at that setting delivered the best results. So while I would meter, rate, and expose for a different ISO depending on the scene at hand, I would process at ISO 320. I think that may be the source of some of my confusion. That, and the fact that false color does not match what my meter says at the given ISOs...

What I am am talking about is a "dailies" or "offline" look - to convey what I think the image should look like.

When conforming, say, DPX's - the goal would be to process the footage with as much information as possible. Or if you are conforming from the original R3D files, you and the colorist will wheel the ISO up or down as you see fit when you are grading.

 

[Bill Anderson]

This seems to be one of those rare cases of "easier done than said".

OK, how about a RED spot meter that reflects digital standards
and jives with RED spaces? ( : One that displays the displacement
of mid gray as ratings are changed
*----v----* *------v--* . Where '*' is the user defined limits of noise
and highlight detail. Or am I behind the times on this?

 

[Ryan E. Walters]

What I am am talking about is a "dailies" or "offline" look - to convey what I think the image should look like.

When conforming, say, DPX's - the goal would be to process the footage with as much information as possible. Or if you are conforming from the original R3D files, you and the colorist will wheel the ISO up or down as you see fit when you are grading.

I'm following you now, thanks.

 

[Sanjin Jukic]

What I am am talking about is a "dailies" or "offline" look - to convey what I think the image should look like.

When conforming, say, DPX's - the goal would be to process the footage with as much information as possible. Or if you are conforming from the original R3D files, you and the colorist will wheel the ISO up or down as you see fit when you are grading.

grading or CC from R3D files is much easier than the same from DPX Log "cooked" files.

Simply get RR and try to do it yourself in Color, it's a cheapest way to check it out or another is to go with Scratch+RR=REDRocketFuel....

 

[Ryan E. Walters]

Another question / thought-

According to my understanding of how digital sensors are read and store information, under exposing an image is a BAD thing in digital. The first HALF of information (or bits) goes to the first stop of the image, Half of what information (bits) remains goes to the next stop, then half again, and so on. So if I am under exposing my image, I am using less of the potential information (bits) that is available to me and I am not taking advantage of everything that is there.

So wouldn't rating the camera slower (On my meter only), and thus opening up the exposure be better for me? (False Color will read correctly, I'll be using all of the bits that I have available to me, etc.) I of course have to watch my highlights, as once they are gone there is no recovery for them.

 

[Ryan E. Walters]

Simply get RR and try to do it yourself in Color

Red Rocket is not supported by Color yet. Color can grade off of the R3D's, but it it my understanding that RR does not boost the performance of Color yet ...

 

[Sanjin Jukic]

RED1 set @ 320 ISO, lens Leica Summilux-R 50mm f/1.4 @ f/5.6

the light is LED light set @ 500W maximum,

and also you can see at the short movie that play with the lens aperture

one stop up and down, also focus rack + - @ 5.6 aperture, etc...

RED1 False Color captured via HDSDI preview out to AJA Hi5-3G HDSDI to HDMI converter then to BM Intensity Pro card with Log & Capture @ FCP 7: The skin tones are at the PINK color...

3cP Cinematographer's chart 709-2 with detail of the frame in center that was 1:1 magnified by RED EVF and then HDSDI preview out converted to HDMI and captured from R1.

Leica Summilux-R 50mm f/1.4 with aperture @ 5.6

Lighting stage is with LED light @ maximim 500W positioned close to the 3cP chart.

Download here the short movie>>>

that shows zoomed and magnified image of the movie

captured from R1 HDSDI preview out via AJA HDSDI to HDMI to BM Intensity Pro HDMI capturecard

@ FCP 7 timeline that all means @ 709rec.

 

[Tai Wah Lim]

The Third Man Sanjin, pardon my ignorance, what's is the point in showing this shot? If we set to to ISO320 and light accordingly, we will be able to nail the exposure you shot. The question is the disparity between having meter reading at 320 and setting ISO in RED to 320.Or am I missing the point this thread is all about? Lim

 

[Sanjin Jukic]

The Third Man Sanjin, pardon my ignorance, what's is the point in showing this shot? If we set to to ISO320 and light accordingly, we will be able to nail the exposure you shot. The question is the disparity between having meter reading at 320 and setting ISO in RED to 320.Or am I missing the point this thread is all about? Lim

All the story is not to "judge" your external exposure meter (Sekonic, Gossen, digital/analog or what else,...)

just because IT SAID that you should "deeply" trust to RED1 inner built-in exposure meter and to FOLLOW Graeame Nattress

exposure instruction with ETTR, then RED1 False Color instruction and then FINALLY you can NAIL your exposure

to have a head-room to do later with RAW R3D file what ever you want to do.

Also don't forget to think about "equipment calibration inaccuracies or process variables" all the time:


"Exposure index

Exposure index, or EI, refers to speed rating assigned to a particular film and shooting situation in variance to the film's actual speed.

It is used to compensate for equipment calibration inaccuracies or process variables, or to achieve certain effects.

The exposure index may simply be called the speed setting, as compared to the speed rating."


Hope that helps.

 

[Dan Hudgins]

Digital Cinema vs. Video exposure

Digital Cinema vs. Video exposure

In video camera to video tape you have limited dynamic range because of video tape clipping or broadcast clipping so you set the gray steps chart and adjust the 'scope waveforms to get 90% white near the top of the wave form.

35mm ECN has several stops range above 90% white, even more if pull processed.

So Digital Cinema comes along and you get 12bits, you NEED 9 bits in the end, so you have two or three bits to under-expose with and get more film like results, IF the camera is true 100% RAW recording.

That puts 90% white card at about 25% high on the 'Scope for full scale reading. And gives you TWO stops more speed than the way old video cameras were adjusted.

Then comes RED ONE at "12bit" compressed camera. So its got macro noise in the shadows that a true RAW Digital Cinema camera would not, so you can maybe not under-expose as much as you would like.

So you are told to give ONE stop or so to have highlight head room, therefore the camera seems to under-expose faces, where as it is just giving you enough under-exposure to save the specular reflections on the skin.

In video you powder people so they DON'T reflect, in Cinema peope can be as shinny as they like because film can take it. With RED ONE, you make a little under exposure and you take a little clipping. There is NO right answer to these questions, you open and the shadows look better you close the iris and the highlights look better, wait for 14 bits and you have two more stops to relax the exposure settings with...

 

[Sanjin Jukic]

My last quote on this topic that is from my signature:

"If you know your exposures and know what you're doing while shooting,

then it is fundamental to know what you can do with the material afterwards."

Anthony Dod Mantle, BSC, DFF,

Cinematographer, the ASC and Academy Award® Winner on his experience with RED1.

"American Cinematographer" Nov 2009, page 71>>>

 

[Daniel Browning]

According to my understanding of how digital sensors are read and store information, under exposing an image is a BAD thing in digital.

I concur.

The first HALF of information (or bits) goes to the first stop of the image, Half of what information (bits) remains goes to the next stop, then half again, and so on. So if I am under exposing my image, I am using less of the potential information (bits) that is available to me and I am not taking advantage of everything that is there.

I kindly disagree. What you are saying is only true about computer-generated images (CGI). In cameras, the potential information is always limited by noise. Never by the number of bits. The only time that bits plays has a negative effect is when you have posterization (i.e. quantization error). But no self-respecting raw camera engineer would allow that to occur.

So wouldn't rating the camera slower (On my meter only), and thus opening up the exposure be better for me?

The increased number of bits will have no effect, since the camera is (and will always be) limited by noise. But it will decrease the amount of noise; that's where the real benefit of ETTR is.

I of course have to watch my highlights, as once they are gone there is no recovery for them.

Right. The optimal exposure is always the brightest exposure possible without blowing the highlights you care about.

Essentially what matters is light, not bits. :smile:

 

[Florian Stadler]

I kindly disagree. What you are saying is only true about computer-generated images

I kindly disagree that bit depth has nothing to do with live action acquisition. Recording Bit depth is a limiting factor of dynamic range. As a rough rule of thumb, a 12 bit camera will always have less than 12 stops of DR, a 14 bit camera less than 14 and so on.

If your tolerance of noise level is 100%, you can consider a 12 bit camera a 12 stop DR camera....

 

[Daniel Browning]

I kindly disagree that bit depth has nothing to do with live action acquisition. Recording Bit depth is a limiting factor of dynamic range.

No. It is only a limiting factor when the manufacturer of the raw camera was allows quantization error to limit dynamic range instead of noise. While it's theoretically possible, in practice it does not occur. (The Pentax K10D is one exception, but the quantization error exists only at tones that are already corrupted by pattern noise, so it's more properly said that the dynamic range was limited by pattern noise, not quantization error.)

Again, it is noise, not bit depth, that limits dynamic range. Say you have 12 bits and 11.3 stops of dynamic range. If you go from 12-bits to 20 bits without reducing noise, then your dynamic range has not improved at all. You might as well throw away the extra 8 bits and save space. If you reduce noise without increasing bit depth, then you get quantization error.

To focus on bit depth is to miss what's really important. Increasing bit depth is like buying a t-shirt that says "I climbed Mt Everest." That's the easy part. Anyone can do that. Just as anyone can buy a 21-bit ADC off the shelf and put it in their camera (and some manufacturers have). Just as anyone can take a camera that only needs 11 or 12-bits max and start storing 16-bit files (e.g. some Medium Format backs). But that doesn't make it real or useful (unless you want to just lie to people, err.. "marketing").

The hard part is actually climbing the mountain. Then the t-shirt will actually be true. That means actually reducing noise. Only then will you need to do the easy part: higher bit depth.

As a rough rule of thumb, a 12 bit camera will always have less than 12 stops of DR, a 14 bit camera less than 14 and so on.

Agreed. I would only add a clarification that here were are talking about DR per pixel. It's also possible to measure DR in ways that more directly relate to what is seen by the viewer, such as the spatial frequency corresponding to the viewer's CoC. For example, downsampling 4K to 480p increases dynamic range by over 2 stops due to a similar reduction in spatial frequency. As another example, the phase one backs truncated to 12-bits have far more dynamic range than the 14-bit Canon 40D.

If your tolerance of noise level is 100%, you can consider a 12 bit camera a 12 stop DR camera....

That's only if you assume that the manufacturer wearing the t-shirt ("I climbed Everest") is telling the truth. Maybe they did climb Everest, and the camera does have noise levels that are low enough to warrant 12 bits or 14 bits or whatever bit depth they're selling. In practice that is not the case. Marketing departments have found they can just buy the t-shirt without climbing the mountain. Put "14-bits!" on the box even if the last two bits are completely random and a waste of space. The result is that most cameras have higher bit depth than they can possibly use (e.g. Canon 14-bit).

More detail on this subject here.

 

[Tyler Black]

Try a white card reading, use the back of the Kodak Gray card that is 90% white.

This is a perfect example of why I love this site. I've been studying cinematography for over 20 years. The more you study, the greater the gaps become between reading something you never knew before. The reduser group increases my frequency of newly learned things exponentially. I never knew that the back side of a Kodak grey card was a definitive value that can be used for charting. Excellent. Thanks Dan.

 

[Ryan E. Walters]

The Third Man Sanjin, pardon my ignorance, what's is the point in showing this shot? If we set to to ISO320 and light accordingly, we will be able to nail the exposure you shot. The question is the disparity between having meter reading at 320 and setting ISO in RED to 320.Or am I missing the point this thread is all about? Lim

Nope, you are not missing the point at all. Here is my take away from all of this-

1. If your meter is set to ISO 320 and you light the scene accordingly you will be under exposing by about 1 stop. This additional stop will help protect your highlights since you are underexposing the image.

2. If your meter is set to ISO 160 and you light the scene accordingly your meter will now match what the Red is telling you via False Color. (Green and pink now fall where they should.) However, you have now lost the highlight protection you were getting before, and you need to carefully pay attention to your highlights.

Other thoughts to add to the mix based off of replies in this thread and links to other threads:

1. If you are shooting a High Dynamic Range Image, then you should change the ISO in the camera to a higher one & meter accordingly. This will force you to underexpose the image. Resulting in an image that will protect your highlights, but will move the shadows toward the noise floor. If you process the image according to the set ISO, you will be fine*. But if you try and bring up the shadows noise will be introduced into the image as you are trying to push the shadows farther then they want to go. So if you are needing to protect your shadows to allow for more pushing in post then light the shadows accordingly.

2. If you are shooting a Low Dynamic Range Image, then you should change the ISO in the camera to a lower one & meter accordingly. This will force you to overexpose the image. Resulting in an image that will protect your shadows, but will move your highlights towards clipping. If you process the image according to the set ISO, you will be fine*. But if you are trying to recover highlight information you will be unsuccessful, as they have most likely clipped. So if you will be trying to pull more out of your highlights in post, then you will need to protect the highlights on set accordingly.

3. For me, the source of the biggest amount of confusion has been the use of ISO and trying to find one ISO for the camera. After reading and rereading, and testing and retesting, I have come to the place that many others have already come to- I think using the term EI (Exposure Index) is much more helpful. There is no one way to rate the camera (unfortunately). If you rate the camera at ISO 160, when you get to an HDR scene you will be screwed. If you rate the camera at ISO 640, when you get to a LDR scene you will be screwed. If you are shooting an average scene and rate the camera at ISO 320 you will most likely be fine. So I am abandoning the ISO thinking and moving toward the EI system in my thinking. When I am in an HDR setup, I'll think in terms of an EI 500 - 800. When I am in a LDR setup, I'll think in terms of an EI 125 - 200. And for a mid range / average setup I'll think in terms of an EI 250 - 400.

*According to my reading it appears that some of the post processing programs needlessly clip information in the highlights. (There is info there, but they cut it off.) For these programs, it is better to lower the ISO in the program and then grade it back to the shot ISO.

 

[Ryan E. Walters]

Question about post processing-

Question about post processing-

As I have read in various other threads, processing in certain programs needlessly clips the highlight information. (RedCine for example.) But if you lower the ISO of the R3D file and then grade the image you can use all of the info in the highlights.

What then, is the proper way to test and see if a program is clipping the highlights? I want to be able to figure out when a program is doing this, and at what level it is doing it. I'm currently using Red Alert, Red Cine, and Color- but I'd like to know how to do it in general, rather then a "just set the ISO to 200" answer. (Teach me how to fish ...)

 

[Ryan E. Walters]

Right. The optimal exposure is always the brightest exposure possible without blowing the highlights you care about.

Essentially what matters is light, not bits. :smile:

Thanks for the clarification- that makes sense.

 

[Dan Hudgins]

Bits and Noise

Bits and Noise

Somehow people here seem to think that one pixel on the sensor equals one pixel in the processed TIF etc output.

That is not the case.

To derive good chroma from the sensor data you must average from 9 to 1024 or more pixels, using a method that finds the edges of chroma regions. All advanced de-Bayer methods use pixel averaging to reduce the noise in a particular processed pixel.

You should NEVER get 100% noise from the ADC for any bit. If you balance the noise signal from many pixels it comes out to a low value, but the signal MIXED with the noise remains. In fact the dithering from the so called noise helps reduce posterization in the images that the steps in the ADC cause. The sensor signal is analog mixed with various noise sources and therefore is not limited to the bits in the ADC. If you average many pixels you can get MORE bits of data out than the bits in the ADC in a true RAW camera.

The RED ONE is more complex since much of the fine data is removed by the compression so when you average after de-compression you do not get as close to the true values of the light over small areas.

When you start with at 12bit 4K true RAW image and reduce that the right way to DVD size you end up with more than 14bits of useful data.

Even the luma gets interpolated with several pixels so no single pixel in the end result comes from a single pixel in the sensor. The OLPF is used to blur the light over four pixel groups so you get one RED, one BLUE, and two GREEN pixels giving luma signal for any "single" pixel area of the image, you then sharpen (OLPF comp) the image using about 9 pixels to get the luma for each output pixel. So the single pixel "noise" rating is not what the image is made of since you use 9 or more pixels to get the data out with more detail and less dithering from the "random" noise it is MIXED WITH.

Again in the RED ONE, the sensor data is pre-filtered and compressed so the fine detail is removed from the signal and other false signals are mixed into the data from the wavelets so the amount of true data you can later extract is more limited than in a true RAW camera, but starting with 4K helps since the eye and film print are closer to 1280x720 from a normal viewing distance.

The more you blur the images to combine data from several pixels the less the noise in any given pixel matters, and one would hope that there IS some data in the pixel data mixed with bi-polar random noise that can be averaged out.

In a three chip camera you might be able to use single pixel data, but that would not give the highest useful EI so cameras are not built that way any longer, all modern cameras use multi-pixel processing to increase useful EI and to reduce highlight clipping and correct chroma issues.

 

[Daniel Browning]

What then, is the proper way to test and see if a program is clipping the highlights? I want to be able to figure out when a program is doing this, and at what level it is doing it.

I suggest just comparing it to a known-good raw converter as you are now. As you know, one can't assume that it's the manufacturer's own converter, because Canon, for example, doesn't even use the correct white point for most analog gain settings on their own cameras, clipping up to 1/3 stop of highlights in DPP for no reason at all.

When you start with at 12bit 4K true RAW image and reduce that the right way to DVD size you end up with more than 14bits of useful data.

Agreed.