I think I know why Canon sensors lagged behind Sony's...

stsva wrote in post #15065897
He was counting the complaints in parallel universes.

I feel like i'm in one right now.
The universe where you can't move past the OCD guard who's counting the black and white floor tiles...

It's a slow afternoon so I though of another issue. You would also need a separate shutter for each pixel. ISO does not control the amount of light hitting the sensor such as changing shutter speed or aperture, but controls the amount of amplification applied to the data when it is steamed off. So, lets take the most simple example, two pixels focusing on a target pure black on one side and pure white on the other (0 and 255). Further lets assume that the size of the target is such that the entire black 1/2 fills one pixel and the white half the other pixel.

So, I focus and take the picture. Pixel #1 sees all black so it's meter tells it to up the ISO all the way trying to render the black pixel an 18% gray, and pixel #2 seeing all white does the same. Now, which pixel set the aperture and shutter speed? The black pixel because it needs more light to get anything to register? Or the white pixel because an exposure of over a certain setting will blow out that pixel.

Lastly, what is the result? Not more DR but less as the two pixel compete to become 18% gray. I don't see how this is going to give real photographers more DR. Al it will do is compress the image ultimately giving less DR.

I don't know why I am defending the idea but I like it. (Edit: Well actually defending it is more fun than arguing about hyperbole and the use of the word "literally". And because I'm a geek? )

You don't need a separate shutter fox each pixel. Same shutter speed, same aperture. The only variable is ISO.

Think of a GND filter, it causes part of an image to be exposed by N stops of light less than another part of an image, given the same aperture and shutter speed. The result is an image that is more detailed in both the bright and dark parts.

The auto ISO would need to work much in the same way. It knows that Tv and Av MUST be constant. Therefore, the scene can be metered for the brightest part and subsequent zones can be determined to be 1, 2, or N stops darker than the brightest part. Those zones (or as per the initial idea/suggestion, they would be pixels), would be adjusted in ISO by that many stops.

I'm not sure how shooting a 2 pixel black and white scene makes any sense, because you can get a decent shot of a chessboard and get both black and white in it

The issue involves distance and natural lighting, and I don't think can be simplified to black and white (no pun intended).

andrikos wrote in post #15065823
Should I guide you to the definition of "hyperbole for effect"?

No, but you could provide some basis for your unsubstatiated speculation.

You actually don't need a separate shutter or even any changes to iso. Your exposure is whatever it takes to espose the dark areas (how far into that you go may depend on iso settings or just some sort of user preference).

The bright areas will clip out eventually but the only important information is how fast that pixel blows out. Brighter equals faster and you end up with a steeper curve. That just gets interpreted in the software somewhere so that everything is given a relative brightness value.

gjl711 wrote in post #15065953
This got me thinking. A better way would be to double the amount of luminescence levels, that is go from 256 to 512 (or more) This would allow for a serious increase in DR without the need of auto ISO on a per pixel level.

No - doubling the number of levels is just one stop. In my view, a serious increase in DR would require several stops - 3 stops extra would require 8 times as many levels measured. 4 stops would require 16 times as many levels.

In the end, what we would like to have, isn't any auto-ISO. we would like to have non-linear sensors. Either sensors that reports the logarithm so twice as many captured photons isn't measured as a twice as large value. Or sensors that may bleed away excess photons creating some form of shoulder, where the sensors works as normal for dark/medium gray but then as more and more stops of light are pouring in the photosites leaks a larger and larger percentage to make sure that they don't reach that clipping 100%.

ddk632 wrote in post #15066061
The auto ISO would need to work much in the same way. It knows that Tv and Av MUST be constant. Therefore, the scene can be metered for the brightest part and subsequent zones can be determined to be 1, 2, or N stops darker than the brightest part. Those zones (or as per the initial idea/suggestion, they would be pixels), would be adjusted in ISO by that many stops.

When is this process (sense and adjust) taking place, and can it keep up with 12fps and a scene that warrants the use of 12fps?

waterrockets wrote in post #15066108
When is this process (sense and adjust) taking place, and can it keep up with 12fps and a scene that warrants the use of 12fps?

You can't really sense and adjust. If you sense/adjust for individual pixels, you get a gray image. If you sense/adjust for a group of pixels you get artifacts where a strong white part of the measured region will require dimming down of that region thereby also dimming down the darker parts of the region.

In a video camera, this can work since there can be a huge amount of tricks that our eyes aren't fast enough to see. But in photography, we have the time to look at the details. And we regularly crop images and adjust curves. So artifacts of zoned measurements would be visible.

waterrockets wrote in post #15066108
When is this process (sense and adjust) taking place, and can it keep up with 12fps and a scene that warrants the use of 12fps?

No technology is perfect right from the start.

12fps is just a matter of processing power and a shutterspeed of faster than 1/12 of a second (plus mirror tavel time etc). If the scene is too dark for that then you just bump the iso like you do now.

This is all just speculation along the lines of "wouldn't it be great if..."

Just have fun with it.

waterrockets wrote in post #15066108
When is this process (sense and adjust) taking place, and can it keep up with 12fps and a scene that warrants the use of 12fps?

With a powerful enough dedicated metering processor (Digic N; you get the point) 12 fps is nothing. You're talking about microprocessor calculations per second vs. mechanical shutter frames per second.

Microprocessor is going to win that fight

I think that process would take place at the same time the sensor is being exposed to light as it records data in memory, prior to any writing to the CF or SD media taking place.

ddk632 wrote in post #15066061
....
Think of a GND filter, it causes part of an image to be exposed by N stops of light less than another part of an image, given the same aperture and shutter speed. The result is an image that is more detailed in both the bright and dark parts.

The GND analogy does not work because it is physically restricting the number of photons hitting the sensor. In essence, it has temporarily changed the aperture (sort of) for some set of pixels.

ddk632 wrote in post #15066061
The auto ISO would need to work much in the same way. It knows that Tv and Av MUST be constant. Therefore, the scene can be metered for the brightest part and subsequent zones can be determined to be 1, 2, or N stops darker than the brightest part. Those zones (or as per the initial idea/suggestion, they would be pixels), would be adjusted in ISO by that many stops.

This is where the whole concept falls apart. ISO does not physically restrict photons from hitting the sensor as a change in shutter or aperture does. It just changes the level of amplification used. So once a sensel is full, it's full. Reducing it's ISO will not change the value

ddk632 wrote in post #15066061
I'm not sure how shooting a 2 pixel black and white scene makes any sense, because you can get a decent shot of a chessboard and get both black and white in it .

Because it reduces the problem to it's simplest form. Two pixels, can't get simpler than that and a two color target, also as simple as it gets. How would an auto-iso two pixel sensor capture that image maxing out DR, 0 on one end and 255 on the other?

ddk632 wrote in post #15066150
With a powerful enough dedicated metering processor (Digic N; you get the point) 12 fps is nothing. You're talking about microprocessor calculations per second vs. mechanical shutter frames per second.

Microprocessor is going to win that fight

I think that process would take place at the same time the sensor is being exposed to light as it records data in memory, prior to any writing to the CF or SD media taking place.

I agree that processing will eventually be capable of this kind of power, but it's quite a long way off.

Along the lines of zoned processing mentioned above, and totally gray images, I'm not sure that per-pixel ISO makes sense. This is a good conversation though.

It seems to me that we're going to really just need to have more sensitive sensels as technology advances, producing more RGB depth.

[QUOTE=gjl711;15066170​]This is where the whole concept falls apart. ISO does not physically restrict photons from hitting the sensor as a change in shutter or aperture does. It just changes the level of amplification used. So once a sensel is full, it's full. Reducing it's ISO will not change the value
[QUOTE]

The trick is measuring how fast it gets full.

Unregistered.Coward wrote in post #15066063
No, but you could provide some basis for your unsubstatiated speculation.

Did you read the patent?
Come back in a couple of days with your notes...

pwm2 wrote in post #15066104
No - doubling the number of levels is just one stop. In my view, a serious increase in DR would require several stops - 3 stops extra would require 8 times as many levels measured. 4 stops would require 16 times as many levels.

In the end, what we would like to have, isn't any auto-ISO. we would like to have non-linear sensors. Either sensors that reports the logarithm so twice as many captured photons isn't measured as a twice as large value. Or sensors that may bleed away excess photons creating some form of shoulder, where the sensors works as normal for dark/medium gray but then as more and more stops of light are pouring in the photosites leaks a larger and larger percentage to make sure that they don't reach that clipping 100%.

Unfortunately, photocells are linear devices.
But, like you suggested, you can "reset" them after they're "full" reading a light value, lower the ISO and repeat.
You can use regressive algorithms to do this very fast on a pixel level and in effect apply a "custom" GND filter on each pixel.

To think of this in a simple way start with, say, 16 pixels and then keep doubling them until you reach the size of the sensor.

Something like this was impossible a short while ago, but with enough processing power you can do anything.

I can't wait to read about all the complaints about only having enough battery power for 300 shots...