ISO Obsolete?

[Eisen Feuer]

Fixed ISO/ASA/Filmspeedwhatever values served us well when our sensors were... a high priced roll of film. I have to wonder though, in the age of metadata and the DSMC, what is the purpose of keeping to the old world of fixed ISO values?


My argument is in favor of a Variable ISO system, but I can forsee where there might be roadbumps:

1) The sensors designed to work at certain sensitivities and nothing in-between

2) The differences between ISO values are so minute there's no reason to have anything in-between (why we don't care about variable fstop values, presumably)

3) The postprocessing software (especially concerning HDR photography) would need an update at the same time



Discuss- do you see any advantages in the current fixed/stepped ISO value system, or is it only tradition keeping us away from more options in terms of sensor sensitivity?


This isn't meant to incite war, I'm more curious as to whether the modern camera sensor could do such a thing (or is it mostly the A/D converter after the sensor that changes ISO?)

 

[David Mullen ASC]

If you mean why can't you pick values between 50, 64, 80, 100, 125, 160, 200, 250, 320, 400, 500, 640, 800, etc.... well, those are 1/3-stop increments and there isn't a lot of practical value of having ISO ratings in-between such small increments of exposure / sensitivity.

 

[Eisen Feuer]

Aha, see I'm used to 100/200/400/800/1600 etc,

I guess my defeat is in option #2 then.



...


and ignorance.

 

[Tom Lowe]

Very incremental control over ISO/ASA could indeed be useful, IMO. Ramping exposure for timelapse day to night, for instance.

Is there anything preventing a company like Red from making their sensitivity adjustable in tiny increments, like ISO 100, 105, 110, 115... and so on?

 

[Eisen Feuer]

Aha! leave it to the patron saint of timelapses

But really- could you imagine the kind of amazing effects one could do if Filmspeed, Shutter Speed and Aperture were completely variable? ...and programmable?

I'd really like to see the effect of going from an open (circular) aperture to a polygon variably over time and see what happens to starred lens flares, or perform an exposure ramping over time that doesn't effect DoF or shutter speed as per Tom's curiosity above.

But I'm not talking about more incremental, I'm talking about... basically, float values. As variable as an analogue volume knob.

 

[David Mullen ASC]

Very incremental control over ISO/ASA could indeed be useful, IMO. Ramping exposure for timelapse day to night, for instance.

Is there anything preventing a company like Red from making their sensitivity adjustable in tiny increments, like ISO 100, 105, 110, 115... and so on?

250, 320, 400, 500 -- these are 1/3-increments, which is usually the smallest adjustment most people make in exposure. Smaller adjustments than that are usually pretty darn subtle.

Besides, it would be more visually consistent, noise-wise, that each frame exposure in time-lapse use shutter speed or f-stop to make small adjustments in exposure, rather increasing and decreasing the amount of sensor under and overexposure, which is basically what changing the ASA rating would be unless analog gain is being applied. In other words, sure, a RED camera could allow 100, then 105, then 110 ASA on each successive frame... but since that is just metadata, it's not actually making any adjustments at all in exposure!

 

[Dan Hudgins]

Shutter vs. ISO

Shutter vs. ISO

For timelapse it would be good to have micro adjustments to the exposure time so you can hold a peek on the histogram still when the light varies.

From what has been said, the RED ONE has only ONE ISO because the sensor amps gain before the ADC is not adjustable, like it is on some other cameras.

They should gamma adjust the image data before compression (dependent on the ISO selected) to reduce noise in under-exposure but it seems they don't. If they want linear for the de-Bayer they could run anti-gamma after de-compression and there by get linear at various ISO ratings and less "macro" noise from the compression artifacts due to REDCODE.

Small ISO/Exposure changes might not matter with ECN, but on a linear sensor 1/3 stop moves the highlights quite a bit. Smaller changes can help adjust the peaks in the histogram so that you optimize the exposure since Digital Sensors are more like color slide film than color negative film, with reversal when you over-expose there is no density, but with color negative film you can pull-process and over-expose several stops. There is not true pull process for Digital Sensors, minimum gain is full scale on the ADC any less gain would not increase the dynamic range.

 

[David Mullen ASC]

You're not limited to 1/3-stop adjustments when using an f-stop or shutter to control exposure. My point is that when making continuous exposure adjustment on each frame exposure, whether due to speed ramping or long-term time lapse photography, you'd want to make real exposure adjustments, i.e. with the f-stop, shutter, and ultimately, ND filters... not by changing the ISO rating on each frame, which is not a true form of exposure adjustment unless gain is being applied to the signal before recording. Which is not necessarily the best way of doing per-frame exposure changes.

You don't want a time-lapse sequence to basically go from the equivalent of -12 db gain to +12 db gain -- basically 4-stops under to 4-stops over, or an 8-stop range. The noise change would be significant just to get a sequence within a range that the f-stop and shutter time could handle better, and maintain the same noise.

 

[Gavin Greenwalt]

For time lapse it seems like you wouldn't want to touch ISO at all. You'll just be discarding data. If you must then the best way would be to just animate levels in your grading application or run it through something like Furnace DeFlicker in which case on the shoot your only real objective would be to get the broadest, best range of data.

They should gamma adjust the image data before compression (dependent on the ISO selected) to reduce noise in under-exposure but it seems they don't. If they want linear for the de-Bayer they could run anti-gamma after de-compression and there by get linear at various ISO ratings and less "macro" noise from the compression artifacts due to REDCODE.

Pretty sure they do. If you look at the raw data it's very heavily processed prior to compression. It's definitely not just stored in 3 linear channels.

 

[Dan Hudgins]

Fixed gain

Fixed gain

Pretty sure they do. If you look at the raw data it's very heavily processed prior to compression. It's definitely not just stored in 3 linear channels.

They have said many times as far I can understand that the DO just have three linear channels of fixed gain and the ISO --ONLY-- changes the meta-data for later gamma up of the mid-tones and such.

If they have changed the way the camera works, maybe someone who knows can say what is going on now. You can test this by shooting at ISO 250 and ISO 500 with the SAME exposure, then override the ISO for de-Bayer and see if the results look any different (assuming that the override is effective as I think they have said you can override the camera settings.)

For timelapse you DO need to set the ISO for the minimum exposure level that the shot will need, and more or less leave it fixed, then adjust the shutter or iris or variable ND (crossed pola that roatate by motor drive) to have the subject stay within the data clip levels. If you don't have the ISO high enough to start with then at some point in the shot you will be under-esposed, or if it is not low enough and the light goes up you will be over-exposed. With a fixed gain camera like the RED ONE, you need to just think of it as ISO 320 (all the time at 5500K) and use that to base the exposure on, because it WILL clip if you use other numbers that are just meta-data as the light goes up or down without shutter or iris or ND adjustments to keep the light within the sensor's dynamic range.

With film the head and foot are curves and so the contrast varies with exposure, if you rate higher or lower you get various contrast effects, but with digital the sensor is linear so you do not change the contrast with exposure changes, you just clip the top or fall into noise at the bottom, the contrast is constant unless you put an S-curve on in processing the RAW data, but even so the RAW data needs to be on the straght line part of the sensor ADC output.

 

[Mohammed El Sharqawy]

Very incremental control over ISO/ASA could indeed be useful, IMO. Ramping exposure for timelapse day to night, for instance.

Is there anything preventing a company like Red from making their sensitivity adjustable in tiny increments, like ISO 100, 105, 110, 115... and so on?

btw, I wish to ask you what is the best way to adjust exposure for a day to night timelapse..??

 

[Tom Lowe]

With the the 5d2 I think you really have about 5 stops gain with iso raw. Because the downsample to 1080p is so massive, it's usable up to 1600 or 3200. If I could ramp iso maybe 4 or five stops, from say 100 to 1600 that would be helpful for sunset to evening shots, for example.

As David said, shutter speed is the easiest thing, though you will see changes in speed for clouds, etc.

Maybe Red could give us very incremental control over shutter? Like 1 second, 1.1 seconds, 1.2 seconds...

 

[Mohammed El Sharqawy]

my cameras doesn't enjoy all that res,, I have nikon D5000, Canon 400D and Canon 450D..
isn't there any computer software that can do this ramping in shutter speeds or aperture automatically?

 

[Peter Moretti]

RAW and variable ISO seem to be opposed to each other.

RAW is a capture of minimally processed sensor data. Variable ISO would need the ability to apply varying amounts of gain to the RAW data, thus making it "less RAW."

Red's theory of operation is that its 12-bit linear data gives enough resolution that this type of image manipulation can (and should) be done in post.

Some argue that 12 bits is not enough and that much more expensive competitors use higher levels of resolution for their in-camera basic image adjustments. I don't know how significant this is.

But I believe there is more than one way to alter gain/vary a camera's ISO, and, BTW, some of these methods do not entail amplifying sensor data. One is to change the gamma encoding curve used.

Another is to remap the levels at which analog to digital conversion takes place to apply more bits of resolution to where there is useful image information. For example, if the composition is very dark, most the the bits of resolution is wasted on mid and highlight levels that do not exist in the composition. It's kind of like using a foot long ruler to measure a group of objects, none of which is longer than three inches; you have 12 inches of measurement, but nine of those inches go wasted.

Let's say the ruler is incremented by 1/8th's of inch. Imagine you could shrink the ruler to half its size. Now you'd have a ruler only six inches long with 1/16th's increments. This would be an improvement for measuring the objects b/c even the longest is no more than three inches long, and the very small ones will be measured with greater precision. You have doubled the resolution and lost nothing. You could even shrink the size of the ruler by one half again and gain (no pun intended) even more resolution. Now the ruler would be three inches long with 1/32nd's increments. The precision has doubled again and the largest object can still be measured.

This is like remapping the D/A conversion levels of the sensor. You do this by changing the electrical value assigned to one bit. Half the value and you have twice the resolution but also half the latitude. But dark compositions don't require wide latitude, they require greater accuracy in the shadows.

This is how I understand the process to work on some sensors and I hope I have explained it properly.

 

[Gavin Greenwalt]

They have said many times as far I can understand that the DO just have three linear channels of fixed gain and the ISO --ONLY-- changes the meta-data for later gamma up of the mid-tones and such.

Wait... what? I was responding to image compression. What were you talking about I must have been confused.

Edit: Oh I get what you were saying. I missed the gamma-based-on-iso comment. Ok well they do gamma encode the channels before compression it seems but it's not iso-dependent.

 

[Eisen Feuer]

Peter, that is an excellent idea in terms of changing the gamma curve so bits don't go to waste in dark compositions. The compression codec could better do its job in preserving the integrity of dark situations, and all you'd need is metadata to keep the data dark when viewed (because compressing the bits like that, when viewed without adjustment would make the whole scene look much brighter and fake). There is the question of signal:noise, and exactly how said gamma curve is implemented. It may be a hardware thing, or as simple as firmware. I wonder what the engineers have to say about the feasibility of such a thing, or if it's something that's already been tested?

 

[Dan Hudgins]

Non-Linear

Non-Linear

Where did you read that they apply gamma before compression, from what I have read they seem to use linear data into compression?

To do the non-linear ADC thing, you can use a log-amp between the sensor data and the ADC.

The more usual way to is us linear ADC then a 1D LUT to convert the 14bit linear to 10bit log.

If you program the LUT with gamma curves that change with the ISO you can optimize the mid-tone gradation.

Camcorders that save H.264 MUST use gamma or hyper-gamma curves before the compression since H.264 has less range than REDCODE. But adding ISO dependent pre-compression gamma should help improve the results since the JPEG2000 like compression used does show aritifacts in under-exposed parts of the frame.

 

[Peter Moretti]

Peter, that is an excellent idea in terms of changing the gamma curve so bits don't go to waste in dark compositions. The compression codec could better do its job in preserving the integrity of dark situations, and all you'd need is metadata to keep the data dark when viewed (because compressing the bits like that, when viewed without adjustment would make the whole scene look much brighter and fake). There is the question of signal:noise, and exactly how said gamma curve is implemented. It may be a hardware thing, or as simple as firmware. I wonder what the engineers have to say about the feasibility of such a thing, or if it's something that's already been tested?

Eisen,

I'd love to take credit for the ideas in what I wrote, but I was actually explaining how I understand some digital movie cameras do image processing in regards to light sensitivity.

There are two separate approaches. 1. The use and "shapability" of the gamma curve. 2. Scaling the sensor's A/D conversion; I believe this approach is only used w/ CCD not CMOS chips.

 

[Eisen Feuer]

Never heard of such a thing being implemented, my experience has always been Dark + Compression = SOL

 

[Peter Moretti]

Well selectable gamma curves have been around for a long time, mostly used to increase latitude and reduce the chances of blowing out the highlights.

And how the camera adds gain is not something that the user will be aware of. But it will affect how much gain you can get away w/ adding before the image starts to look like crap.