External Monitors and Scaling IRE Range

[Jon Betz]

I am looking to better understand an issue I am having with my monitor and inform myself about future monitor choices.

I own a SmallHD 702 Bright and Dragon-X 6K. When I send the monitor an image from the Red, the IRE range is not scaled correctly (zero is not zero and clipping is not at 100). I can adjust the scaling in the monitor settings, but this is quite cumbersome and also changes with different ISO's and IPP2 settings. Can someone explain what's happening there? Is this a limitation of the SmallHD monitors, and does it also exist on the newer units (like their larger 13", 17" and larger monitors)?

Do other monitors handle the Red image better and automatically scale the input to correctly display the range of values from zero to 100?

Thanks much.

 

[David Rasberry]

Small HD waveform monitors should display the actual IRE/data range of the camera output as recorded or the IRE/data range of the monitor viewing lut applied to it. Rarely if ever will either signal fill the full scale range of the scope nor should it, especially if one is recording scene values with a log curve applied. The IRE range of a recorded log signal is as flat and compressed as it looks on the monitor absent a viewing LUT.

 

[Jon Betz]

Small HD waveform monitors should display the actual IRE/data range of the camera output as recorded or the IRE/data range of the monitor viewing lut applied to it. Rarely if ever will either signal fill the full scale range of the scope nor should it, especially if one is recording scene values with a log curve applied. The IRE range of a recorded log signal is as flat and compressed as it looks on the monitor absent a viewing LUT.

Edit: I am referring to IRE values of the monitor out with the 709 monitor LUT applied in the Red, not anything to do with log.

So false color on SmallHD is completely useless? If I set false color to a specific range for skin tones but clipping occurs at 90 IRE, the false color range I have set is useless for correct exposure. And if I change ISO the output range changes. This isn’t the case on other monitors correct? I believe I read something at one point about automatic scaling in some monitors.

I guess I understand showing an exact representation of the output range in theory but that doesn’t really make any sense from an actual use perspective. For example I created a false color preset on my SmallHD that uses the same IRE ranges and colors as the RED Video false color. Except unless I go in and change that pre/post setting on the SmallHD every time I change ISO, the monitor false color is constantly wrong and does not match the RED Video setting.

Does Flanders do this?

 

[David Rasberry]

The clip point of a viewing lut is determined by the color space it represents. Black clip for 10 bit Rec709 is 7.5 IRE. White clip is 90.
In 10 bit code that corresponds to 64 black and 940 white. But luts don’t indicate the clip point of the camera sensor. The scopes only indicate relative visual value of your exposures under that particular lut. REC709 at 2.4 gamma maps a 10.5 stop value range between black clip and white clip.

 

[David Rasberry]

There is an option setting shown in the manual for the SHD monitors for the scopes to show camera data values instead of the display LUT values. Don't know if that applies to false color or not.
Except for Gemini, ISO changes don't affect the clip points of the sensor at either end. Just the relative position of middle gray. Frankly I think the histogram is more useful for raw exposure judgement.

 

[RyanFrancis]

I think the false color on flanders is fantastic. I've actually not used it on my smallHDs.

If I'm understanding you correctly, different ISO's, ipp2 ranges, LUTS, will DEFINITELY affect your false color ranges. As I understand it, false color is a tool for checking your IRE values POST luts.

 

[David Rasberry]

I think the false color on flanders is fantastic. I've actually not used it on my smallHDs.

If I'm understanding you correctly, different ISO's, ipp2 ranges, LUTS, will DEFINITELY affect your false color ranges. As I understand it, false color is a tool for checking your IRE values POST luts.

That's how it works on my Bolex too. ISO changes only affect what you see in the monitor path, not what the sensor exposure values are. There is an undebayered raw luminance view option on the Bolex that lets you see exactly what is clipped and what is not at the sensor level right down to the pixel when zoomed in. That is a useful exposure tool. Apparently RED has a similar tool, though I have not heard any discussion of it. You would not use it while shooting, just as a raw exposure check. The Bolex reverts to either REC709 or Bolex log/wide gamut in the monitor path, whichever is selected, as soon as you push record.
Mostly I expose by an incident light meter, just like film. Scopes and false color are for video cameras, though I could see the usefulness of them if shooting in Prores to a display color space standard. Raw, not so much.

 

[Phil Holland]

"Video Check" in the RED ecosystem is False Color and it does take into account your IPP2 transform, LUT, or anything you are doing to the image. A little tricky there as IPP2 being the new kid on the block is actually designed not to clip.

"GIOSCOPE" is the sexiest tool RED has created to stare deeply into the sensor's raw feed to discern what is a clean image, what is a noisy image, what is underexposed, and what is clipping. It's painfully useful and I keep it hot keyed on my DSMC2 cameras as I do a fair bit of technical filming and like to max out my DR often.

 

[Jon Betz]

Thanks Phil. I was hoping to do some test of the different IPP2 contrast and highlight settings to accompany my reply but I have not had time to put that together yet.

It seems there are two things going on here:

1. The sensor has an inherent sensitivity and therefore "black" and "white" point, by which I mean a minimum amount of light below which no photons will be registered by the photo sites, and a maximum amount of light above which everything will appear white (clipped). Obviously in general terms we want to keep our image between these bounds.

2. The IPP2 monitor transform changes how the raw sensor data is output to monitors by applying curves. In my testing, both the contrast adjustment and highlight rolloff adjustment change how wide the range of IRE values are on a 709 monitor. The highlight rolloff adjustment also changes the steepness of the curve in the highlights, so affects both the "softness" of the highlight rolloff as well as how wide that IRE range is on an external monitor.

In my past experience, I have tended to expose skin tones on red around 55-65 IRE depending on complexion (using RED's 709 transforms, not RLF). In recent years I have gotten more accustomed to using false color to get there (either on RED's monitors using the Video false color or using false color modes on a SmallHD or Flanders).

Presumably, the sensor best renders skin tones within a particular range of light values hitting the sensor. For the sake of argument let's just say that's at 800 ISO and with a 709 curve applied, around 60 IRE for Caucasian skin. This would be very good info to apply on set. But if the IPP2 monitor transform settings change the way the RLF image is spread across the 709 space in different ways, 60IRE with one combo of settings represents a different amount of light hitting the sensor than 60 IRE with a different combo of IPP2 settings. And none of the available settings spread the image across the full range of 0-100 IRE. You can clearly see where the image clips on the 709 image and it is never at 100 IRE. If I take the resulting clips with IPP2 applied into Davinci or any other software and expand the range so that black is 0 and clipping is at 100, the skin tone value I chose on set changes and I am not getting what I intended. Isn't this problematic? I guess the fact is, a lot of lighting and skin tone exposure ends up being part science and part intuition and part just looking at the monitor on set and trusting you're getting it right. That's fine. But it seems like a more repeatable method should be possible.

To contrast, if I apply a 709 LUT to a Canon camera and output that monitor LUT in-camera to an external monitor, the resulting image always goes from 0 to 100 IRE and the 100 IRE clip point matches the actual clipping point of the log image. So I know that the IRE value of my skin tones is a repeatable thing.

 

[Jake Wilganowski]

GIOSCOPE is the best thing about red cameras

 

[Mark A. Jaeger]

I second or third or xx the GIOscope = Super helpful to set the real exposure.

 

[Michael Tiemann]

Thanks Phil. I was hoping to do some test of the different IPP2 contrast and highlight settings to accompany my reply but I have not had time to put that together yet.

It seems there are two things going on here:

1. The sensor has an inherent sensitivity and therefore "black" and "white" point, by which I mean a minimum amount of light below which no photons will be registered by the photo sites, and a maximum amount of light above which everything will appear white (clipped). Obviously in general terms we want to keep our image between these bounds.

2. The IPP2 monitor transform changes how the raw sensor data is output to monitors by applying curves. In my testing, both the contrast adjustment and highlight rolloff adjustment change how wide the range of IRE values are on a 709 monitor. The highlight rolloff adjustment also changes the steepness of the curve in the highlights, so affects both the "softness" of the highlight rolloff as well as how wide that IRE range is on an external monitor.

In my past experience, I have tended to expose skin tones on red around 55-65 IRE depending on complexion (using RED's 709 transforms, not RLF). In recent years I have gotten more accustomed to using false color to get there (either on RED's monitors using the Video false color or using false color modes on a SmallHD or Flanders).

Presumably, the sensor best renders skin tones within a particular range of light values hitting the sensor. For the sake of argument let's just say that's at 800 ISO and with a 709 curve applied, around 60 IRE for Caucasian skin. This would be very good info to apply on set. But if the IPP2 monitor transform settings change the way the RLF image is spread across the 709 space in different ways, 60IRE with one combo of settings represents a different amount of light hitting the sensor than 60 IRE with a different combo of IPP2 settings. And none of the available settings spread the image across the full range of 0-100 IRE. You can clearly see where the image clips on the 709 image and it is never at 100 IRE. If I take the resulting clips with IPP2 applied into Davinci or any other software and expand the range so that black is 0 and clipping is at 100, the skin tone value I chose on set changes and I am not getting what I intended. Isn't this problematic? I guess the fact is, a lot of lighting and skin tone exposure ends up being part science and part intuition and part just looking at the monitor on set and trusting you're getting it right. That's fine. But it seems like a more repeatable method should be possible.

To contrast, if I apply a 709 LUT to a Canon camera and output that monitor LUT in-camera to an external monitor, the resulting image always goes from 0 to 100 IRE and the 100 IRE clip point matches the actual clipping point of the log image. So I know that the IRE value of my skin tones is a repeatable thing.

It would be helpful to take a step back and consider what tool is asking to do what job. REC 709, as I'm sure you know, maps a fictitious 10-stop range to about 6 stops of display dynamic range. That's what it's designed to do. RED cameras are designed to deliver more than 16 stops of dynamic range. Whether they do that or not is a subject of fun debate, but that's what they are designed to do. The GIO LUT tool was designed to show the RED camera's 16 stops of dynamic range. RED's IRE tool was designed to show how, after various transformations, the native dynamic range of the camera maps to the voltage levels sent to a display (using the IRE scale, not an actual voltage scale). The IRE scope on a monitor is also supposed to tell you those levels being sent to the pixels on the display. My Flanders Scientific DM240 monitor has several levels of LUTs, some of which operate behind the scenes (calibration LUTs and Gamma) and some of which impose on the incoming signal, affecting the values that would then be "measured" by the IRE scope that comes later in the signal chain. I'm sure your SmallHD monitor has a manual that shows its signal flow.

What you need to do is to draw a signal flow diagram that can represent everything from the initial sensor capture, through each conversion you apply (gamma, contrast, lift, gain, offset, slope, contrast, saturation, LUT, etc), to final display and then ask yourself this simple question: am I actually measuring what I care most about, or am I measuring something that's already baked or not even yet baked at all? If you are trying to sort out what's the best value for skin tones, it's not after the LUT. It's way upstream in the sensor data.

And you might also want to understand what output levels cause the LUT to really shine. BlackMagic did something really interesting with their launch of the 12K camera and Gen5 color science. Their REC2020 PQ LUT is >PERFECT< when viewing the images on a FSI SM310K REC2020 HDR monitor with ST2084 PQ gamma. The same raw images going through their REC 709 LUT looks like the same cartoon garbage that made me dump the BMCC 6 years ago for RED DRAGON. I'll leave it to smarter people than myself how it would be possible in 2020 that they cannot produce a working REC 709 LUT for < 10 stops of dynamic range, but there you go.

Phil Holland's LUTs are very specifically designed to work within the RED workflow (translating IPP2 Log3G10/RWGRGB to Log3G10/RWGRGB). They can be used and abused for creative purposes, but as a tool, they have a well-defined place in the workflow. Other LUTs have other functions.

Use the right tool for the job, and work becomes easy and fun.

 

[David Rasberry]

REC 709, as I'm sure you know, maps a fictitious 10-stop range to about 6 stops of display dynamic range. That's what it's designed to do. I'll leave it to smarter people than myself how it would be possible in 2020 that they cannot produce a working REC 709 LUT for < 10 stops of dynamic range, but there you go.
.

REC709 and its predecessor analog NTSC both map a mathematical 10 stop +/- defined photometric dynamic range depending on applied gamma. Any signal values outside that range are clipped. That is independent of the photometric performance of any particular display technology used to view the image. A modern HDR capable LED display can display a true 10 stops from an REC709 input if it has the photometric range to do so. REC709 hybrid log gamma LUTS in Resolve can expand the displayed range to about 11 stops. Higher dynamic range sensor values that one wants to preserve in an image have to be compressed to fit within REC709, I.E. modifying the value curves at both ends to preserve relevant image details and present a more pleasing result. The 10 stop range and the standard defined IRE scope values of 10 IRE= 1 photometric f/stop visually correspond to an Ansel Adams Zone scale.

The origin of both is physics research into human visual response in the 1920's - 1930's that discovered the brain interprets visual input into a 7 to 10 photometric stop representation of a scene at any given instance. Our photographic systems all represent the power density value distribution curve of human vision to one extent or another. Otherwise it doesn't look natural to us.

The fundamental equation for matching electronic photo systems voltage values to human visual response is: 20log(white clip^gamma/black clip^gamma)= dynamic range in dB/6= DR in photometric f/stops. Display reference standards set clip point values relative to a defined non-zero analog voltage range.

NTSC standard values in IRE: black clip is 7.5 IRE, white 110 IRE corresponding to 47mv black-700mv white at gamma 2.5. In practice NTSC analog monitor gamma values ranged between 2.35 and 2.6.

REC709 defined values for digital encoding 16 black clip/235 white clip for 8 bit and 64 black clip/940 white clip for 10 bit over a typical range of 2.4-2.6 gamma Note that digital encoding standards apply a scale of defined discrete non-zero numeric values to a defined analog voltage range. REC709 references NTSC voltage values.

At gamma 2.5 the resulting DR for each are:

NTSC analog voltage = 58.65 dB = 9.77 f/stops

NTSC IRE= 58.32 dB = 9.72 f/stops

8-bit REC709 = 58.35 dB =9.72 f/stops

10 bit REC709 = 58.35 = 9.72 f/stops

NTSC related visual systems I am intimately familiar with as an old video engineer. My physics professor at Furman was Tom Goldsmith, worked for DuMont labs in the 1930's developing the physics and technology that ultimately resulted in the NTSC color standard. He was one of the key patent holders and headed the committee that implemented the standard.

I'm not fully up to speed on the newer standards yet. But all digital display reference standards are still tied to a defined range of analog values of some kind.

Linear encoding of raw sensor data in digital cinema cameras is similar but is a 10 log function over the analog voltage range of the sensor. I can illustrate this with the 12 bit CDNG values that my Bolex encodes from a truncated 16 bit first stage AD conversion, Black clip is 8 and white clip is 3750, representing the analog voltage noise floor to full well white clip range. 2.6 gamma conversion referenced to REC709 HLG in Resolve color managed workflow captures the full usable DR of the camera from linear XYZ input color space.

10log(3750^2.6/8^2.6)=69.44dB=11.57 f/stops. This is an accurate representation of the maximum DR of the camera as tested visually.