D600 RAW is not much better than A900

Started Sep 27, 2012 | Discussions
ck3
ck3 Senior Member • Posts: 2,925
Re: D600 RAW is not much better than A900

shady1991 wrote:

It's a known fact that a900 a700 didn't have NR off setting in their raw files in its first firmware. you can read everywhere about that even in dpr.

I corrected that sentence for you.

shoevarek Contributing Member • Posts: 518
How many bought A900 with the intention of frequently using it at ISO3200?

I wonder. The difference in IQ at high ISO setting would need to be quite large, not only regarding noise but also regarding dynamic range and color reproduction, for me to justify the upgrade (assuming other things constant).

shady1991 Regular Member • Posts: 218
Re: D600 RAW is not much better than A900

maybe I mixed up that fact, but it's obvious that in DPR samples of a900 NR is applied.

 shady1991's gear list:shady1991's gear list
Sony SLT-A57
ck3
ck3 Senior Member • Posts: 2,925
Re: D600 RAW is not much better than A900

shady1991 wrote:

maybe I mixed up that fact, but it's obvious that in DPR samples of a900 NR is applied.

You mean: "The reasons which led me to my original conclusion turned out to be false, but I stick to it anyway."

And while we're at it: The NR settings in camera (weak, strong etc.) do not describe the quality of the algorithm but the degree to which NR is applied.

ck3
ck3 Senior Member • Posts: 2,925
Re: D600 RAW is not much better than A900

Amateur Sony Shooter wrote:

DPR has added D600 studio image to their database for us to compare different cameras. When I looked the RAW up to ISO3200, I don't really see meaningful difference between D600, 5D3, D800 and A900. These new generation sensors might be one stop better than A900 in therms of high ISO. Did I miss anything? http://www.dpreview.com/previews/nikon-d600/7

I think you might be a bit too optimistic with regards to improvements in high ISO quality. I'd agree that the D600 has about one stop advantage on the A900 (which unfairly had the reputation of being a noise machine) and the D3X which isn't too shabby. Add to that the (probably) improved DR due to less noise in the shadows and you've got a better product. If the improvements warrant to buy the new model is for each to decide.

GuyMcKie Contributing Member • Posts: 611
Re: D600 RAW is not much better than A900

Firefox supports aRGB.

With "muted" I mean some less saturation in some colors as red and green, compared with 5DIII, and more with the strong saturation of the A900.

D600 raw files looks very good. 25600 iso better than the 5DIII and the D800.
Must be more than one stop better than the A900.

Michel J Veteran Member • Posts: 4,009
Re: D600 RAW is not much better than A900

Amateur Sony Shooter wrote:

DPR has added D600 studio image to their database for us to compare different cameras. When I looked the RAW up to ISO3200, I don't really see meaningful difference between D600, 5D3, D800 and A900. These new generation sensors might be one stop better than A900 in therms of high ISO. Did I miss anything?

Not better, because of the limits of the laws of physics?

Reason why I suggested a dedicated sensor for low ISO here:
http://forums.dpreview.com/forums/read.asp?forum=1014&message=41972223

Kind Regards,

-- hide signature --

Michel J

 Michel J's gear list:Michel J's gear list
Sony SLT-A77 Sony SLT-A37 Konica Minolta Maxxum 7D Sony 70-200mm F2.8 G Sony DT 18-55mm F3.5-5.6 SAM +10 more
TrojMacReady
TrojMacReady Veteran Member • Posts: 8,724
Re: D600 RAW is not much better than A900

Michel J wrote:

Amateur Sony Shooter wrote:

DPR has added D600 studio image to their database for us to compare different cameras. When I looked the RAW up to ISO3200, I don't really see meaningful difference between D600, 5D3, D800 and A900. These new generation sensors might be one stop better than A900 in therms of high ISO. Did I miss anything?

Not better, because of the limits of the laws of physics?

Reason why I suggested a dedicated sensor for low ISO here:
http://forums.dpreview.com/forums/read.asp?forum=1014&message=41972223

Purpose built astrophotography sensors are within reach of the current Exmors, so there's not much to win (in terms of photons for a single exposure time), regardless of design.

Allan Olesen Veteran Member • Posts: 3,387
Re: D600 RAW is not much better than A900

Michel J wrote:

Reason why I suggested a dedicated sensor for low ISO here:
http://forums.dpreview.com/forums/read.asp?forum=1014&message=41972223

I am surprised that nobody in that thread explained to you that a large part of the noise in an image actually comes from the incoming light itself, not from the sensors or electronics. Nothing you do to improve the sensor can remove that part of the noise.

The reason is that at the small quantities of light which hit a sensor, the light cannot anymore be viewed as a homogenous, constant stream of light. It is a stream of single photons, arriving randomly. This randomness creates noise. At ISO 800,000, this noise will be gigantic because only a few photons will hit each sensor pixel.

Michel J Veteran Member • Posts: 4,009
Re: D600 RAW is not much better than A900

Allan Olesen wrote:

Michel J wrote:

Reason why I suggested a dedicated sensor for low ISO here:
http://forums.dpreview.com/forums/read.asp?forum=1014&message=41972223

I am surprised that nobody in that thread explained to you that a large part of the noise in an image actually comes from the incoming light itself, not from the sensors or electronics. Nothing you do to improve the sensor can remove that part of the noise.

The reason is that at the small quantities of light which hit a sensor, the light cannot anymore be viewed as a homogenous, constant stream of light. It is a stream of single photons, arriving randomly. This randomness creates noise. At ISO 800,000, this noise will be gigantic because only a few photons will hit each sensor pixel.

Thanks. Excellent explanation to understand what's happen with photons in a very low light situations...
In the other hand, do nocturnal animals, see random noise at night? I guess: no.

So, what I suggest, is not only a sensor, but an whole system, fully dedicated to low light conditions, with:

  • Backlit sensor to loose a minimum of photons;

  • innovative design of optics that would multiplying the ambient light optically (such as a converging system what can align multiple circles of images through a prism)

  • no more infrared filter to capture more (surnumerary) photons;

  • an in-camera algorithm/processing that would restore infrared images, and interpret their colours to render them such as human eyes perceive it normally;

But I'm not sure if all that is realistic ahah

Kind Regards,

-- hide signature --

Michel J

 Michel J's gear list:Michel J's gear list
Sony SLT-A77 Sony SLT-A37 Konica Minolta Maxxum 7D Sony 70-200mm F2.8 G Sony DT 18-55mm F3.5-5.6 SAM +10 more
Michel J Veteran Member • Posts: 4,009
Re: D600 RAW is not much better than A900

TrojMacReady wrote:

Michel J wrote:

Amateur Sony Shooter wrote:

DPR has added D600 studio image to their database for us to compare different cameras. When I looked the RAW up to ISO3200, I don't really see meaningful difference between D600, 5D3, D800 and A900. These new generation sensors might be one stop better than A900 in therms of high ISO. Did I miss anything?

Not better, because of the limits of the laws of physics?

Reason why I suggested a dedicated sensor for low ISO here:
http://forums.dpreview.com/forums/read.asp?forum=1014&message=41972223

Purpose built astrophotography sensors are within reach of the current Exmors, so there's not much to win (in terms of photons for a single exposure time), regardless of design.

Thanks. Yes you right (and I did not know).

But why not a FF sensor with the backlit design?

Sony told «it was not necessary according to the large surface of photodiodes in the DSLR world » , so... I'm not sure at all about this argument

Kind Regards,

-- hide signature --

Michel J

 Michel J's gear list:Michel J's gear list
Sony SLT-A77 Sony SLT-A37 Konica Minolta Maxxum 7D Sony 70-200mm F2.8 G Sony DT 18-55mm F3.5-5.6 SAM +10 more
Allan Olesen Veteran Member • Posts: 3,387
Re: D600 RAW is not much better than A900

Michel J wrote:

In the other hand, do nocturnal animals, see random noise at night? I guess: no.

I know that the human brain does a lot of tricks to compensate for the flaws in our vision. I would assume the same happens for those nocturnal animals.

  • Backlit sensor to loose a minimum of photons;

Modern sensors lose less than half of the photons. So not much more to gain there.

  • innovative design of optics that would multiplying the ambient light optically (such as a converging system what can align multiple circles of images through a prism)

Sounds like a complicated way of achieving the same as you would achieve with a larger aperture.

  • no more infrared filter to capture more (surnumerary) photons;

  • an in-camera algorithm/processing that would restore infrared images, and interpret their colours to render them such as human eyes perceive it normally;

With the current technology, the camera can only know that those photons were infrared if it blocks them for some pixels and allows them for other pixels - just like the bayer filter we have today to make color photos. This blocking results in a loss of photons, so you may not gain a net benefit.

On the other hand, if someone invents a sensor which can count all photons and at the same time discern between their wavelengths without blocking some of them in a bayer filter, we would take a step forward in sensitivity. I don't even know if that is theoretically possible (there are some laws of physics which say that you cannot theoretically measure all properties for some systems precisely at the same time).

Michel J Veteran Member • Posts: 4,009
Re: D600 RAW is not much better than A900

Allan Olesen wrote:

Michel J wrote:

In the other hand, do nocturnal animals, see random noise at night? I guess: no.

I know that the human brain does a lot of tricks to compensate for the flaws in our vision. I would assume the same happens for those nocturnal animals.

Yes, you absolutely right. Sometimes, when you close your eyes, you can see (or not) such an effect like that (or something... because could we call that "noise"(?), I have no idea, but it's something we can see in parallel with added to the retinal persistence). And so, we can see (or not) because our brain can choose to select it (or not...)

Current NR algorithms tends to smooth it, but not with an adjustable threshold by photographer, nor according to a better exposure ( ETTR workflow) nor to align the red channel as Iliah Borg suggest (by filtering "high frequencies" VS "slow frequencies" photons)

All above can optimise treatment for low light condition. But it doesn't seems to be completely implemented today. (Imho)

For achieving that: Iliah Borg was using different gradations of magenta filters:

  • Backlit sensor to loose a minimum of photons;

Modern sensors lose less than half of the photons. So not much more to gain there.

Hum....! If «less than 50%», that suggest a big margin to make some progress...

  • innovative design of optics that would multiplying the ambient light optically (such as a converging system what can align multiple circles of images through a prism)

Sounds like a complicated way of achieving the same as you would achieve with a larger aperture.

It's already inside the R&D secret drawers, I guess.
.

  • no more infrared filter to capture more (surnumerary) photons;

  • an in-camera algorithm/processing that would restore infrared images, and interpret their colours to render them such as human eyes perceive it normally;

With the current technology, the camera can only know that those photons were infrared if it blocks them for some pixels and allows them for other pixels - just like the bayer filter we have today to make color photos. This blocking results in a loss of photons, so you may not gain a net benefit.

On the other hand, if someone invents a sensor which can count all photons and at the same time discern between their wavelengths without blocking some of them in a bayer filter, we would take a step forward in sensitivity. I don't even know if that is theoretically possible (there are some laws of physics which say that you cannot theoretically measure all properties for some systems precisely at the same time).

Who talk about Bayer, me?
Have already patents for new design, without Bayer matrix.

But we have also Foveon and Sony version of the Foveon sensor.

And/or, like a combination with both: Foveon and back-illuminated design!

Here the standard ones:

Here the new Sony design (what resolve the Iliah Bork dilemma):

Btw, a new sensor dedicated to lowlights could have a bits count conducive for lowlights
(not for highlights)

From extreme highlights to deepest lowlights, like this:

EV N°1 : 1 bits (extreme highlights)
EV N°2 : 2 bits
EV N°3 : 4 bits
EV N°4 : 8 bits
EV N°5 : 16 bits
EV N°6 : 32 bits
EV N°7 : 64 bits
EV N°8 : 128 bits
EV N°9 : 256 bits
EV N°10 : 512 bits
EV N°11 : 1024 bits
EV N°12 : 2048 bits
EV N°13 : 4096 bits
EV N°14 : 8192 bits (deepest lowlights)

As you can see with such inversed bits count, you have much more gradations available in lowlights areas. Same as when our brain can choose to select something more in deepest lowlights...

Because it seems to be that the current bits count starting up from lowlights and go to highlights (according to a non sinusoidal pattern of grey levels in the digital world, what can blow high lights quickly, but what it dosen't matter when we suffer of a lack of photon in lowlights) and so, they are conducive to highlights (mathematically much more grey levels than lowlights) if you know what mean....

Kind Regards,
--
Michel J

 Michel J's gear list:Michel J's gear list
Sony SLT-A77 Sony SLT-A37 Konica Minolta Maxxum 7D Sony 70-200mm F2.8 G Sony DT 18-55mm F3.5-5.6 SAM +10 more
Allan Olesen Veteran Member • Posts: 3,387
Re: D600 RAW is not much better than A900

Michel J wrote:

Allan Olesen wrote:

Modern sensors lose less than half of the photons. So not much more to gain there.

Hum....! If «less than 50%», that suggest a big margin to make some progress...

That is less than 1 stop. In your other thread, you were claiming much larger improvements.

With the current technology, the camera can only know that those photons were infrared if it blocks them for some pixels and allows them for other pixels - just like the bayer filter we have today to make color photos. This blocking results in a loss of photons, so you may not gain a net benefit.

On the other hand, if someone invents a sensor which can count all photons and at the same time discern between their wavelengths without blocking some of them in a bayer filter, we would take a step forward in sensitivity. I don't even know if that is theoretically possible (there are some laws of physics which say that you cannot theoretically measure all properties for some systems precisely at the same time).

Who talk about Bayer, me?

My point was that the most efficient color sensors we have today use Bayer.

It may very well be that other technologies have the potential to be more efficient because the don't need to throw away photons to discern between colors. That was actually covered in my second paragraph.

But as long as these sensors have not yet proven that ability, I will assume that they have the same problem.

Have already patents for new design, without Bayer matrix.

But we have also Foveon and Sony version of the Foveon sensor.

I don't know enough about the Foveon sensor. I assume that it also has the same disadvantage of having to throw photons away to discern between colors, or some other similar disadvantage. Otherwise I would have expected it to prove better than a normal sensor with Bayer filter years ago.

Btw, a new sensor dedicated to lowlights could have a bits count conducive for lowlights
(not for highlights)

From extreme highlights to deepest lowlights, like this:

EV N°1 : 1 bits (extreme highlights)
EV N°2 : 2 bits
EV N°3 : 4 bits
EV N°4 : 8 bits
EV N°5 : 16 bits
EV N°6 : 32 bits
EV N°7 : 64 bits
EV N°8 : 128 bits
EV N°9 : 256 bits
EV N°10 : 512 bits
EV N°11 : 1024 bits
EV N°12 : 2048 bits
EV N°13 : 4096 bits
EV N°14 : 8192 bits (deepest lowlights)

You are essentially asking for a non-linear A/D-conversion. But what do you achieve by having such fine-grained bit values at low light that one additional photon would increase the pixel value with several 1000? One photon is the lowest possible increase in light quantity. Trying to count photons in quantities of 0.001 is just a waste of bits.

(Modern cameras actually try to count fractions of photons already around ISO 1000. No need to make it even worse.).

thubleau7 Contributing Member • Posts: 516
Re: D600 RAW is not much better than A900

Foveon consists of 3 layers stacked RGB

"The operation of the Foveon X3 sensor is quite different from that of the Bayer filter image sensor more commonly used in digital cameras. In the Bayer sensor, each photosite in the array consists of a single light sensor (either CMOS or CCD) that, as a result of filtration, is exposed to only one of the three primary colors, red, green, or blue. Constructing a full color image from a Bayer sensor requires demosaicing, an interpolative process in which the output pixel associated with each photosite is assigned an RGB value based in part on the level of red, green, and blue reported by those photosites adjacent to it. The Foveon X3 sensor creates its RGB color output for each photosite by combining the outputs of each of the stacked photodiodes at each of its photosites. This operational difference results in several significant consequences."

Allan Olesen Veteran Member • Posts: 3,387
Re: D600 RAW is not much better than A900

thubleau7 wrote:

Foveon consists of 3 layers stacked RGB

"The operation of the Foveon X3 sensor is quite different from that of the Bayer filter image sensor more commonly used in digital cameras. In the Bayer sensor, each photosite in the array consists of a single light sensor (either CMOS or CCD) that, as a result of filtration, is exposed to only one of the three primary colors, red, green, or blue. Constructing a full color image from a Bayer sensor requires demosaicing, an interpolative process in which the output pixel associated with each photosite is assigned an RGB value based in part on the level of red, green, and blue reported by those photosites adjacent to it. The Foveon X3 sensor creates its RGB color output for each photosite by combining the outputs of each of the stacked photodiodes at each of its photosites. This operational difference results in several significant consequences."

That did not tell me anything I did not know. And most important:

It did not tell me if a Foveon sensor needs to throw away photons to discern between colors.

TrojMacReady
TrojMacReady Veteran Member • Posts: 8,724
Re: D600 RAW is not much better than A900

If we simplify things a bit (or a lot...), imagine the electronics being a fixed size (as small as reasonably possible), thus blocking a fixed sized portion of the light reaching a photodiode. The smaller the photodiode, the larger the relative portion of light being blocked, the more loss of photons per fixed unit area.

Again, greatly simplified, but that's largely the reason why small pixel backlit sensors gain back much more light per fixed unit area than large pixel sensors. For 1/2.3" 12 MP sensors that can mean up to a stop diference in light or noise. Keeping those tiny electronics constant, that's much less than 1/6th of a stop for current FF sensors. Which doesn't warrant the extra costs at this point. By the time we see well over 100MP FF sensors, you'll probably see this technique making its way into such large sensors as well, because the pixels will be small enough to benefit visibly from it.

Fraxinus excelsior Regular Member • Posts: 348
Re: D600 RAW is not much better than A900

Amateur Sony Shooter wrote:

DPR has added D600 studio image to their database for us to compare different cameras. When I looked the RAW up to ISO3200, I don't really see meaningful difference between D600, 5D3, D800 and A900. These new generation sensors might be one stop better than A900 in therms of high ISO. Did I miss anything? http://www.dpreview.com/previews/nikon-d600/7

As a a900 owner I am maybe a bit colored oppinion wise but nontheless I honestly think the a900 are missing a stop at max. Ofcourse as we go over iso 6400 the a900 cant compete because its only reaching as far as iso 6400.

I think I will jump an generation and my wallet/wife thinks thats a good idea too;-)

D

Michel J Veteran Member • Posts: 4,009
Re: D600 RAW is not much better than A900

Thanks for all your previous comments Allan.

Allan Olesen wrote:

Michel J wrote:

Allan Olesen wrote:

Btw, a new sensor dedicated to lowlights could have a bits count conducive for lowlights
(not for highlights)

From extreme highlights to deepest lowlights, like this:

EV N°1 : 1 bits (extreme highlights)
EV N°2 : 2 bits
EV N°3 : 4 bits
EV N°4 : 8 bits
EV N°5 : 16 bits
EV N°6 : 32 bits
EV N°7 : 64 bits
EV N°8 : 128 bits
EV N°9 : 256 bits
EV N°10 : 512 bits
EV N°11 : 1024 bits
EV N°12 : 2048 bits
EV N°13 : 4096 bits
EV N°14 : 8192 bits (deepest lowlights)

You are essentially asking for a non-linear A/D-conversion. But what do you achieve by having such fine-grained bit values at low light that one additional photon would increase the pixel value with several 1000? One photon is the lowest possible increase in light quantity. Trying to count photons in quantities of 0.001 is just a waste of bits.

(Modern cameras actually try to count fractions of photons already around ISO 1000. No need to make it even worse.).

No, I don't asking for a non-linear A/D-conversion. But if that can help to achieve better lowlights, why not. The actual design is not conducive for lowlights. Period. As a reprographer, I saw many time the lack of details in lowlights area (from DSLR), comparing to a Dainippon scan of MF slides (or very best 35mm). What tell to me we have a loss of some details here, maybe from this bits count, the linear A/D-conversion or watever: but we can expect more from a camera dedicated to lowlights imho, if so....

In the other hand, I can't imagine that only one innovation can provide more gain. Its a combined development to get that, I guess...

Modern sensors lose less than half of the photons. So not much more to gain there.

Hum....! If «less than 50%», that suggest a big margin to make some progress...

That is less than 1 stop. In your other thread, you were claiming much larger improvements.

I was absolutely wrong to this point. I've corrected my overly optimistic conclusions in the dedicated thread.

With the current technology, the camera can only know that those photons were infrared if it blocks them for some pixels and allows them for other pixels - just like the bayer filter we have today to make color photos. This blocking results in a loss of photons, so you may not gain a net benefit.

On the other hand, if someone invents a sensor which can count all photons and at the same time discern between their wavelengths without blocking some of them in a bayer filter, we would take a step forward in sensitivity. I don't even know if that is theoretically possible (there are some laws of physics which say that you cannot theoretically measure all properties for some systems precisely at the same time).

Who talk about Bayer, me?

My point was that the most efficient color sensors we have today use Bayer.

It may very well be that other technologies have the potential to be more efficient because the don't need to throw away photons to discern between colors. That was actually covered in my second paragraph.

Why can't we change of paradigm here? The EVF embedded in the DSLR design tell to us that current design would be change to get better optimisation

Moreover, since we know that the EVF can show things indistinguishable in OVF (in lowlights) there will be more demand to justify R&D innovations in this direction..

Here, the sensor don't percieve colours (of course) but frequencies, and they can sort too according to the penetration in the substrate what balancing well the colours according to the size of sensors area dedicated to. A very clever design:

Gain expected: about a half stop and more(?)

But as long as these sensors have not yet proven that ability, I will assume that they have the same problem.

Have already patents for new design, without Bayer matrix.

But we have also Foveon and Sony version of the Foveon sensor.

I don't know enough about the Foveon sensor. I assume that it also has the same disadvantage of having to throw photons away to discern between colors, or some other similar disadvantage. Otherwise I would have expected it to prove better than a normal sensor with Bayer filter years ago.

Are you pessimist? I notice that you don't react about the idea of Iliah Borg?

Kind Regards,

-- hide signature --

Michel J

 Michel J's gear list:Michel J's gear list
Sony SLT-A77 Sony SLT-A37 Konica Minolta Maxxum 7D Sony 70-200mm F2.8 G Sony DT 18-55mm F3.5-5.6 SAM +10 more
OP Amateur Sony Shooter Veteran Member • Posts: 5,417
Re: D600 RAW is not much better than A900

The thing is, from studio image comparison point of view, there isn't much clear difference between new sensor and A900. Most of the improvement is in jpeg processing and host of other features (video, etc.). The very hard core RAW performance is not much improved that much at all. Besides, who use ISO 6400 anyway? I have used ISO 1600 in rare occasions, maybe a few times 3200 for "testing" and NR exercise.

 Amateur Sony Shooter's gear list:Amateur Sony Shooter's gear list
Sony RX100 Sony a77 II Sony 70-200mm F2.8 G Sigma 10-20mm F4-5.6 EX DC HSM Sony 24-70mm F2.8 ZA SSM Carl Zeiss Vario-Sonnar T* +5 more
Keyboard shortcuts:
FForum MMy threads