Native ISO Question

[Matthew Cernohorsky]

I understand that the Red One's sensor records what it sees from its native ISO, and then you can interpret that footage with different metadata in post, but I don't understand what actually happens when you use an ISO below the native ISO in RedCine-X (I know it's not recommended, but still). If you had clipped highlights at native ISO, wouldn't using a lower ISO just make them a uniform grey? Also, what defines a clipped highlight in RAW. Is that when something is clipped at the native ISO? I guess those questions are very similar. Thanks for your time!

 

[David Mullen ASC]

Clipped just means that bright detail has been burned out to white and thus has no texture or information in the whites. If you overexpose the sensor more by using a lower ISO, then more bright detail will get clipped, i.e. you will reach the point of clipping sooner. The sensor has a set sensitivity and dynamic range; if no information is recorded in RAW because it is too dark or too bright, then there's no information there to play with later in post. Think of RAW as basically everything the sensor received and was able to record; anything after that point can only be less than what was captured in RAW.

Now if you look at the image processed for an ISO selection, then the RAW image is being transformed into something else -- for example, if you select a higher ISO rating, the RAW image will be processed to become brighter, in which case some bright areas of the frame will get even brighter and may look clipped. But if you go back and look at the RAW file, you'd see that there was more information in the highlights before it got boosted by the higher ISO rating. However it doesn't necessarily work that way the other way around -- if you pick a low ISO rating and see something clipped, that means it is also clipped in RAW. However, at the other end, sometimes by darkening the RAW image to create the lower ISO image, shadow detail may be getting crushed darker while on the RAW file, you'd see more shadow detail.

 

[James Drake]

I was going to reply but you got your answer from the best

 

[Matthew Cernohorsky]

Thanks both of you! that makes a lot of sense but I'm still not quite understanding what clipped in RAW means. I understand the concept and what it actually is, but does it mean that the image is clipped at the native ISO?

 

[David Mullen ASC]

Forget the native ISO business. Clipped in RAW means that it is clipped, period, now and forever. The ISO is irrelevant -- if it's RAW then there is no ISO applied yet. It's clipped because it was too bright for the sensor to handle. You have "x" luminance range that can be captured and whatever is clipped fell outside of that.

 

[Gunleik Groven]

I understand that the Red One's sensor records what it sees from its native ISO, and then you can interpret that footage with different metadata in post, but I don't understand what actually happens when you use an ISO below the native ISO in RedCine-X (I know it's not recommended, but still). If you had clipped highlights at native ISO, wouldn't using a lower ISO just make them a uniform grey? Also, what defines a clipped highlight in RAW. Is that when something is clipped at the native ISO? I guess those questions are very similar. Thanks for your time!

There are no good reason to SHOOT with lower than ISO 320, there are a few valid post reasons and scenarios where a lower iso in post may help getting clipping back into safe-zone.

 

[Dan Hudgins]

DRX

DRX

wouldn't using a lower ISO just make them a uniform grey?

The three colors in the sensor do not clip at the same EI/ISO for any given K value of the light at the time of exposure.

The RAW data from the ADC values looks greenish, because the EI/ISO of the green pixels is about half to 2/3 stop higher than the red and blue, the red is about 1/4 stop faster than the blue and more so if ND filters are uses because they counter the far red cutoff of the IR cut filter.

Because the luma can be taken from a mix of the red, green, and blue, you get a soft-clip of the luma when DRX is used to do such a blend.

The chroma is hard clipped so that you do not get color shifts, but when mixed with luma, you do not notice the chroma being hard clipped much because there is still detail in it.

If the green clips at EI/ISO 160 then the blue will probably hard clip at about EI/ISO 100 and the red at about EI/ISO 120 under daylight.

Under tungsten the blue will have lower EI/ISO than the RED so you get a wider DRX spread.

If the highlight subject mater is light in color as it would be, then the lower quality of chroma information is not very noticeable and you can get an extra 1/2 to 2/3 stop of exposure (lower EI/ISO numbers) and still see all the patches on a gray step chart since such a chart is not a saturated color subject.

Having the colors unbalanced in the sensor helps reduce some de-Bayer artifacts and gives an over-all higher EI/ISO so it is more of a benefit to the dynamic range and "film like look" of the images. There is some color shift if you expose above the green clip since the ratios of green to red/blue are not quite right, but most bright subjects are low saturation and you do not notice it so much on them.

For normal subjects you want at least two stops above 90% white card in the green pixels, so you should be shooting at EI/ISO 640 rather than 160 as the maximum exposure value.

If you have 12bits, then you go down 2 bits (stops) to get head room above 90% white, that gives you 10 bits of linear data, then you apply the 0.45 gamma correction and black lift in the sensor so you end up with about 8bits of histogram gap free data at about EI/ISO 640.

Any EI/ISO higher than about 640 will be giving you less than about 8 bits of data, although the noise and large pixel area will downsample to HD to give about +2 bits going from 5K to 1920x1080. That makes 6 bits at 5K or a maximum useful EI/ISO of 2560 (when the sensor data is saved 12 bit linear). The gamma 0.45 correction makes the shadow tones spread so that is where the gaps show up the most, if you "crush the blacks" then you can hide some of those gaps by shifting down the lower mid tones making them darker than they would be with full correction.

How exactly the 16bit modes work in EPIC I have not seen explained, it would seem they log encode them as 12bit data for the REDCODE storage?

If you log encode 16 bits to 12 bits you improve the shadow banding, but unless you change the log curve for the EI/ISO shot, you reduce the highlight data so if you then try to shot at lower EI/ISO you get more banding in the highlights. RED seems to use fixed encoding for the REDCODE, so any log encoding would be a compromise between highlight and shadow bit spacing.

 

[nigelsmith]

Hi Gunleik.
What about if you are working with available light, have runout of NDs, and need your lens wide open for shallow DOF?
I've had students shoot 100EI on Red one M for this reason and the results seem ok.

 

[Gunleik Groven]

Hi Gunleik.
What about if you are working with available light, have runout of NDs, and need your lens wide open for shallow DOF?
I've had students shoot 100EI on Red one M for this reason and the results seem ok.

It doesn't really matter as long as the sensor is the same. Both R1 M and R1 MX are calibrated at 320 (that is not the same as saying they ARE 320...) and shooting below that is only asking for unwanted problems.

Shooting above and posting below can make a lot of sense, though...

If I may refer to my own discussion of the subject:
http://bit.ly/mCwcoN

Cheers!

G

 

[David Mullen ASC]

Hi Gunleik.
What about if you are working with available light, have runout of NDs, and need your lens wide open for shallow DOF?
I've had students shoot 100EI on Red one M for this reason and the results seem ok.

You've probably lost about 1.5-stops of headroom (and gained it in the shadows) which isn't always visible -- especially in less sunny, more northern locations like London...

Of course, you could expose to avoid clipping in really bright situations, but then you are defacto rating the camera a bit faster, for example, if you set the camera to 100 ISO and then close-down 1-stop to protect some hot highlight, you are defacto treating the camera at 200 ISO.

 

[Blair S. Paulsen]

Meta-musings (not Metamucil ;-)

Meta-musings (not Metamucil ;-)

The ISO you set on the RedOne or Epic does 2 things:
1) Changes how "bright" the image appears in the monitor path and what you see in the histogram on the LCD/EVF (there are some caveats here but let's keep this simple).
2) Changes how "bright" the image appears when opened in RCX (or most other software) based on the ISO carried in the metadata.

The ISO you set on camera does NOT:
add any actual gain to the recorded RAW image data
change the actual sensitivity or dynamic range of the sensor

FWIW on R1MX or Epic I typically keep the ISO at 800 on camera so I am less likely to open the iris too far on set. Works for me, YMMV.

As soon as I have the decoded RGB image up on a calibrated monitor in a proper viewing environment (and perhaps even on a scope) the first thing I do is reset the WB (color temp in Kelvin) using a picker and my eye.
The second thing I do is reset the ISO in Scratch (or RCX, Storm, etc) to land the mid-greys in a place I like - sometimes its as low as 320, though 500 or 640 are usually where I end up.
I'd tell you what I do after that but then I'd have to bill you ;-)

In terms of the students who reset the ISO on camera to 100 when lacking NDs and had good results that is due to the inherent DR of the sensor - the amount of actual clipping would have been the same had they set the ISO to 3200 because its metadata.

If perhaps you don't fully grasp what metadata means I suggest Wiki. With r3d data files this means that many on camera settings are not actually changing the image data created by the camera, merely telling the program you are opening the files in what the DP was seeing and potentially intending. Of course some camera settings are very "real" and can't be adjusted later: focus, iris (stop), shutter angle (speed), frame rate (yes, retiming can be done but that's a different thing) and clipped audio levels.

The way some production to post-production pipelines operate the metadata values get "baked in" along the way (during a "dumb" transcode to ProRes for example). If this is your world, then dialing in the metadata takes on a different meaning and function for you. Some people meticulously adjust all kinds of parameters like ISO, WB, tint, etc since they know it will be baked in. Done carefully, this can get you fairly close to the look you intended but it leaves a great deal of the power of a RAW workflow on the table and constrains image quality potential. HTH.

Cheers - #19

 

[David Mullen ASC]

In terms of the students who reset the ISO on camera to 100 when lacking NDs and had good results that is due to the inherent DR of the sensor - the amount of actual clipping would have been the same had they set the ISO to 3200 because its metadata.

Well, not if you close down by five stops once you see how it looks set to 3200 ASA instead of 100 ASA, then the clipping point is going to change, it's just that the total dynamic range is the same, it's just weighted/shifted to a different area.

 

[Gunleik Groven]

Well, not if you close down by five stops once you see how it looks set to 3200 ASA instead of 100 ASA, then the clipping point is going to change, it's just that the total dynamic range is the same, it's just weighted/shifted to a different area.

This is very important to grasp and often messed up!

 

[nigelsmith]

Well, not if you close down by five stops once you see how it looks set to 3200 ASA instead of 100 ASA, then the clipping point is going to change, it's just that the total dynamic range is the same, it's just weighted/shifted to a different area.

Thanks for all the responses guys, particularly David. What you said here matches with my understanding and what I tell students - although ISO IS metadata, it affects monitoring which will often affect exposure decisions. And iris, focus and shutter speed are definitely NOT metadata!
Nigel

 

[nigelsmith]

If I may refer to my own discussion of the subject:
http://bit.ly/mCwcoN

Cheers!

G

Good article Gunleik.
Thanks for the link
Nigel