320-ISO 5000 K stuff

[amourgues]

This "camera always recording at 320-ISO and 5000 K" stuff gets me confused.
Can anybody tell me in simple words what the hell does it mean ?
Doesn´t it matter what iso you choose ?
You just fix the ISO and adjust the iris opening to get the right exposure ?
what ISO to shoot with then ?
what ISO / K settings for different light environments ? (sunny day. cloudy day, interiors at day/night, exterior nights ?)

I know there aren´t straight answers, I know it always dependes...but some help would be really appreciated

thanks a lot

 

[Brandon Fraley]

the sensor is natively 5000k and rated at 320 iso. You can change iso and color temp, but you're stretching the sensor farther away from it's native position. does that make sense? Basically shoot to taste. if it looks good then it is good Just understand that if you find yourself getting too much noise, you'll probably need to add more light and shoot with bluer light.

 

[Darren Orange]

Or look at it like a car....my car natively natively requires 93 octane gas. If I put the lesser gas in like lets say 87, it may still run and likely will but not as well as it would on 93. Anytime you move away from the native sensor rateing( for example going to 250 or 500ISO or to 6000K or 4000K), you can view it as lowering your octane.

**please note that lowering octane in a car can break it....last i checked giving no light to RED ONE will not break it but you may not get any images

 

[amourgues]

Thanks guys
Let me get this straight....Are you suggesting to shoot 320-ISO, 5000 K ?? always ??...no matter what the light setting is ?..then adjust iris and light setting so to get the right exposure ?

 

[amourgues]

I´m more confused now...As I had understtod so far (read elsewhere), irrespective of the user settings, the sensor records at 320 ISO and 5000 K always....don´t get it

 

[Lauri Kettunen]

Amourgues, although the ISO and color temperature are metadata, the sensor has some properties that are fixed. That is, although you changed the ISO and color temperature, this does not affect the signal from the sensor which is stored on CF cards or Red drives. Or, in other words, you are not able to set the physical properties of the sensor by changing the settings. Instead, only the physical properties dictate how the sensor "reads" light. Still, saving ISO and color temperature as metadata yields far more latitude in post than fixing them while recording.

 

[Ryan Manes]

The sensor doesn't record at any ISO on its own. You can set the ISO to whatever you want. But to get the cleanest noise free Image you want to try and shoot at ISO 320.
Now that might not always be possible, say if you were shooting outside at night and didn't have enough lights to get good exposure at 320, you might have to crank the ISO to say 500. Think of it as if you were boosting the gain on a video camera. The higher you go the lower the quality.
This is a some what separate issue than The camera natively being balanced at 5000K temperature. 5000 is what the sensor likes the most, it is what will provide the cleanest image. But that does not mean that you can't get great footage using 3200k tungsten. As far as suggesting what to shoot always, I try to always gel my tungsten lights to bring them closer to 5000K, and I always try to shoot at 320 or below. That doesn't mean I always can. Sometimes the nature of the shoot dictates what I have to do and you just have to know what effect adjusting those settings will give you

 

[Lauri Kettunen]

5000 is what the sensor likes the most

... guess this just means, if the ambient light is 5000K and one shoots a white target, then the signals of red, green and blue channels from the sensor are equal/in balance. In other cases one or two channels dominate over the others, and as a result not all channels get stored with the same bit depth. This is to say, not a maximal of data is stored. This 5000K follow from the hardware properties of the sensor.

 

[Roberto Lequeux]

Isn't it that the sensor is "balanced" for 5000k meaning 5000k light looks white on the RAW file, and "rated" at 320ISO as in the RAW file looks properly exposed when lighting is proper for 320ISO film shot with equal exposure and apperture? And that the reason why 5000k is less noisy is because it is the "middle" or the sweet spot.

And noise only "becomes visible" when pushing the data far enough. Otherwise the ISO set on the camera would have to be an increase of voltage to the sensor or some other hardware level tweak.

And that anything else you do including setting WB and ASA on the camera is the same as doing it in post and only a digital adjustment to the RAW readout from the sensor?

Either way, what's most important to know is that 5000k and 320ISO are the sweet spots.

 

[Roberto Lequeux]

... guess this just means, if the ambient light is 5000K and one shoots a white target, then the signals of red, green and blue channels from the sensor are equal/in balance. In other cases one or two channels dominate over the others, and as a result not all channels get stored with the same bit depth. This is to say, not a maximal of data is stored. This 5000K follow from the hardware properties of the sensor.

This would make a lot of sense! Only I don't understand mathematically why being off RGB balance would mean a lower bit depth. Please explain.

 

[Daniel Browning]

This "camera always recording at 320-ISO and 5000 K" stuff gets me confused.

RED ONE ISO is like film. You can rate ISO 320 film at any ISO you want. 640, 160, etc. But whatever rating you decide to use, it will not chemically alter the film to make it more or less sensitive. The rating only gets passed to the lab as instructions, such as "push 1 stop" or "pull 1 stop".

ISO on the RED ONE, in the same way, is just instructions for processing.

For most cameras, you have to be careful to get the ISO (gain, curve, etc.) just right in the camera, because quantization and compression will negatively affect the image if it's suboptimal. RED does well no matter what setting you choose, because it's just instructions (metadata).

Color temperature on the RED ONE, similarly, is only instructions for processing. Other cameras have you modify the color balance (e.g. change Daylight to Tungsten) and they will bake that into the image irreversibly.

All cameras have certain permanent characteristics. No setting in the camera can change them. Two of these characteristics are QE (quantum efficiency) and spectral sensitivity.

QE means how much of the light is recorded for a given spectrum of light (usually green). Spectral sensitivity means the response to the rest of the spectrum (blue, red, infrared).

Everyone, even manufacturers, have a lot of leeway in deciding what to recommend as an ISO value for a camera. The most important factors are QE and read noise, but other things like recording medium, processing, and display can affect the rating.

Most cameras have a variable gain control, which doesn't improve the QE of the camera, but is often necessary to prevent quantization, compression artifacts, and can sometimes reduce read noise.

Essentially, it's one of the advantages of raw.

 

[Eki Halkka]

This would make a lot of sense! Only I don't understand mathematically why being off RGB balance would mean a lower bit depth. Please explain.

If, say, green and red channel get twice the exposure as blue channel, blue channel will only use half of the digital values compared to them - it will have one bit less precision. In example:

red = values 0-4095 = 12 bit precision
green = values 0-4095 = 12 bit precision
blue = values 0-2047 = 11 bit precision

Or in other words, each f-stop of underexposure reduces one bit of captured bit depth.

To get the cleanest image out of the camera, one should forget about ISO, and simply always expose the image as bright as possible without blowing out any important detail at the sensor (i hear the false color display is good for this...).

***

I might be wrong, but i'm under the expression that the ISO 320 comes from "expose a gray card correctly for this, and you will have four stops of latitude over it before clipping to white".

If you expose for ISO 640, you will have five stops before clipping, but 6dB more noise and you lose one bit of precision.

If you expose for ISO 160, you will have only three stops before clipping, but 6dB less noise and you gain one bit of precision.

And so on...

 

[Roberto Lequeux]

aha... if the blue channel gets half the exposure then all of the info on the blue channel goes from zero up half way the possible values... and it won't grade as well... perfect! and every time you cut it in half that is equal to losing one bit, phenomenal... love it when stuff makes sense, thanks Eki

14 bit will be too sick!

and thanks Daniel for that explanation

 

[Daniel S Williams]

Yeah, so if a particular channel isn't fully exposed, it's potential amount of brightness values aren't fully exploited, right??

 

[Roberto Lequeux]

yea, like an RGB histogram where there nothing but a little bump that doesn't stretch from 0 to 255

or a luma histogram where you were underexposed and the slope dies out before reaching all the way to the right, or where you had to clip the blacks real hard to avoid noise from say an HVX on an underexposed shot and the little @$%#ing hill doesn't reach all the way to the left :watsup: right?

 

[Daniel S Williams]

Haha yeah man just like that.

But, under say Tungsten lighting, there is no way for the blue channel to be fully exposed, ever....so is that where the Red cam gets noisy? It seems like there are huge number of blue photosites on the sensor though, regardless of the fact it's a Bayer filter... Are 3 CCD cameras better in Tungsten light, usually?

 

[Daniel Browning]

In other cases one or two channels dominate over the others, and as a result not all channels get stored with the same bit depth. This is to say, not a maximal of data is stored.

If, say, green and red channel get twice the exposure as blue channel, blue channel will only use half of the digital values compared to them - it will have one bit less precision. In example:

red = values 0-4095 = 12 bit precision
green = values 0-4095 = 12 bit precision
blue = values 0-2047 = 11 bit precision

Or in other words, each f-stop of underexposure reduces one bit of captured bit depth.

To get the cleanest image out of the camera, one should forget about ISO, and simply always expose the image as bright as possible without blowing out any important detail at the sensor (i hear the false color display is good for this...).

aha... if the blue channel gets half the exposure then all of the info on the blue channel goes from zero up half way the possible values... and it won't grade as well...

Yeah, so if a particular channel isn't fully exposed, it's potential amount of brightness values aren't fully exploited, right?

This is all technically correct. 99% of photographers, videographers, and other people believe that the loss in precision/number of levels is the most important factor in underexposure/ETTR and would have said the same things you did.

Unfortunately, it's also irrelevant, unimportant, and misleading. (I'm one of the 1% that dissents from the commonly held view.)

The reason underexposure causes more noise and problems in post production is *not* because it causes the use of fewer levels or utilization of less of the bit depth: it's because the ratio of signal to noise (S/N) is worse.

If the RED ONE had 22-bits (0-4 million levels, one stop underexposure is 2 million), the quality (noise, tonal gradation, etc.) would be *exactly* the same as now with 11 bits (0-2047 levels for one stop underexposure).

If the RED ONE had 7-bits (0-128 levels), the quality would still be the same except for quantization (which can cause quantization error or quantization "noise"; as well as posterization).

All well designed raw cameras have enough bit depth to prevent quantization at *any* exposure. Even *12 stops* underexposed. So exposure will never cause any problems related to the number of levels. Quantization has never been demonstrated in a raw REDCODE file, so it appears to be just like other well designed raw cameras.

Suggested reading: http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/

Brighter exposure increases S/N; dimmer exposure decreases S/N. Bits are just trivia: it's the amount of light falling on the sensor that is important.

 

[Daniel Browning]

But, under say Tungsten lighting, there is no way for the blue channel to be fully exposed, ever....so is that where the Red cam gets noisy?

I would put it a different way.

Unfiltered tungsten light has half or a third as much blue light. Sensors can't see something that isn't there, so it naturally results in underexposed blue channel. To make matters worse, current sensors are slightly less sensitive to blue, so it's underexposed even more, relative to red and green.

To get the image back to a natural color balance, the blue in pushed in the raw converter, which makes the noise much more visible.

It seems like there are huge number of blue photosites on the sensor though, regardless of the fact it's a Bayer filter... Are 3 CCD cameras better in Tungsten light, usually?

I'm not certain, but I would think the 25% higher percentage of blue-filtered pixels would result in a better image, AOTBE (all other things being equal, such as total sensor area, aperture, etc.)

It would be hard to actually test the case because most 3-chip cameras bake the image a lot inside the camera (black clip, white balance, saturation, noise reduction, gamma, tone curve, etc.).

 

[Roberto Lequeux]

Well yes, I never discounted S/N... if there is noise then you would ruin the image regardless of how many bits you were using to record the signal... not the rest of the stuff you layed out here Daniel... :blink: Whoa dude... just when I thought I knew it all

So I guess the best thing is to use 5000k light and if you want that pleasing tungsten look you can easily do it in post... DO IT IN POST... meh, we'll do it in post... ahahaha... that little expression is going to make a comeback in a way most who hated it never would have imagined :usd:

 

[Lauri Kettunen]

This would make a lot of sense! Only I don't understand mathematically why being off RGB balance would mean a lower bit depth. Please explain.

Although Halsu and others have given answers here's my version:

Assume you shoot "to the right" and exposure just below clipping. If the color temperature of the ambient light is not the nominal temperature of the sensor, then the channel with maximal signal sets the limit to your exposure. The other channels remain below clipping.

Every stop from the maximum means one half of the data is not in use. Or the other way around, every stop doubles the amount of data you have. Since Red One is a 12-bit system, say if your G channel is a stop from the maximum, you have 11-bits of G data. Two stops from the maximum imply 10-bits of G data and so one.

This is why filters are often useful. If the ambient color temperature is far from the nominal value, it is motivated to reduce the channel with the biggest signal in order to increase the amount of data stored on the other channels.

Perhaps should add, the main difference between film and CMOS sensors is that the response of the CMOS sensor is linear, i.e., directly proportional to the amount of light. But our eye is not. For this reason a gamma curve (or more presicely, a gamma function) is needed. The very idea of the raw format is that you can leave the specification of the gamma curve to post. No need to fix it while shooting. Thus, the very goal is only to maximize the amount of data --shoot to the right-- and leave everything else into post.

Since films are typically made to respond the same way as an eye, it's a different ballpark and you don't shoot to the right the same way. If you do, you won't be able to come back to the left the same way. For, the response of film is not directly proportional to the amount of light. Thus, if you took one half of the signal on film this would not correspond to one half of the light exposured. In principle, if films with linear response were made and existed --I have no idea of this-- then workflow of such film and CMOS-sensor were the same.