Thu 07-Jun-12 03:59 AM | edited Thu 07-Jun-12 09:25 AM by nrothschild
>>A sensor's noise is best considered in terms of the amount >of >>noise per square millimeter of sensor. No matter how you >>slice up each millimeter, the noise is essentially the >same. >>In other words, doubling or halving the number of pixels >per >>millimeter will not change the noise in each millimeter >of >>sensor. > >If this is true, then pixel density on the chip is a non >factor. Correct?
Yes, that is exactly what I am saying there. Pixel density has no bearing on noise when noise is measured as a function of unit sensor area.
>>Regardless if you use the TC or not, you end up with the >same >>noise, especially in a very high ISO situation like you >have >>here. > >I agree. What you DO end up with is less light.
I hesitate to parse words here, but technically you do not "lose light", at least relative to the surviving frame when the TC is used. The TC spreads the light out, making it "less dense". The camera meter is reading light density so it sees "a loss of light". The same number of photons hit your soccer player's image on the sensor with or without a TC. But with a 1.4x TC that same quantity of light is spread over twice the sensor surface area.
The light you do lose is the light outside the sensor area when the TC is used. And that too equals half the original light. And in a real sense, that is the lost light the meter "sees" by dropping the exposure 1 stop.
I suspect you understood that, but I have to be clear there.
>>What is somewhat hidden under the covers is that by >shrinking >>the image by 1 stop of focal length, you are increasing >the >>noise when you have to stretch those square millimeters >of >>pixels more in the final output. > >I am confused by this statement. Could you explain what you >mean by "shrinking the images by one stop"? I am >assuming you mean that as we decrease the focal length (in my >presented case from ~420-300mm, we must then increase the >final image by the same amount thus exacerbating the noise >problem. If this is what you mean, then yes I agree.
Yes, that is exactly what I mean and why all else equal DX is a push when it comes to noise. You gain a stop in the exposure and then lose it by magnifying the final output image one stop. Net zero.
>>You have a D800. If you shoot a properly exposed ISO >6400 >>image and then display or print it at some reasonable >output >>size I suspect it probably looks pretty good. Even if it >has >>some challenging tones in it, like those OOF blurs that >can >>look so ugly in a lesser (older) camera. > >This is correct. As we are starting with a LOT more >information, as we shrink the image, the noise shrinks >proportionately. If viewing at 100% this advantage is lost.
If you understood that then you should well understand that in a too low light situation DX is no help at all- in terms of noise. Y0u went to f/2.8 on the front end, gaining a stop, and then lose that stop on the back end when you magnify the sensor sized image to the final output.
>>You are doing the same thing when you drop the TC and go >with >>DX. The image size on each sensor, with the same lens >and >>focal length, is the same size and it contains the exact >same >>amount of noise. > >Disagree. I get less noise because I've gained a stop of >exposure.
You misunderstood me because I could have said it better. What I meant to say just above is that the same lens, same focal length (either both with or both without the TC) will be the same size on the DX and FX sensor. When you drop the TC on the DX sensor, now you have a smaller image that needs to be magnified 1 stop more, nullifying the one stop pickup in aperture and exposure. I was just repeating what I said before and you agreed to.
>>When you drop the TC and go with higher >>density DX, you get the same pixels across your cropped >final >>output > >Also disagree. This ONLY holds if the sensors are the same >pixel density. This would be true if we were talking about a >D3s vs A D300. But in the case of a D400, D7000, etc., this >would NOT be true.
You're splitting hairs. I was implying a D3/D300 comparison for simplicity. A D7000 gives you a slight advantage in resolution over the D3, and surely the D400 will do even better. But the critical point here is that the sensor density has no relationship to the noise, which is your primary problem in the posted image that started this exchange.
>>The result is a net push, or it should be. If it is not, >then >>all the physics behind sensors, and the basic theory of >>"unpredictable photons" (shot noise) needs to >be >>rethought. > >I don't think it's an issue of physics so much as you are >considering certain things as constants that are NOT >constants. They are variables.
We will get to that later.
>>In the final analysis, if all else is equal between the DX >and >>FX sensor > >And this is where the wheels fall off the wagon. All else is >NOT equal. I am not dealing in theory here. I am dealing in >real world cameras that I have in my possession.
One post you are, and the next post you are not, and then you are back again. This thread, like all the others, is primarily about the D400, and you have added the general lack of a fully Pro DX camera with integrated grip, WIFI, hard wired internet ports, etc. - cameras that do not exist. Since they do not exist, we can only assume all else is equal until tests and user experience prove otherwise.
The primary point I am trying to make here is that, conceptually, if all else is equal, meaning the DX and FX sensor is "cut from the same wafer with different size dies", then your plan to go to DX and drop the TC won't work. We have to agree with that, and by extension agree on the basic physics, before we can even begin to talk about specific DX and FX bodies and how they will perform.
Once we agree on that, then if you want to talk about specific bodies you own, you have to convince me that the DX body is inherently less noisy than the D3, meaning it will outperform the D3 in DX crop mode. And I think you will have a tough time finding that DX camera in your bag now. So I'm not sure where you are going with this.
>>(they have the same noise per square millimeter and >>they will if they were designed around the same time) > >Correct, but nothing in our scenario holds to this constant. >The only way I cam get the real world equivalent of this is if >I consider a D3 and a D300. The D7000 is a generation newer, >and the D400 may be another generation beyond that.
Bill Claff's tests suggest the D3s in DX crop mode still beats the D7000 by 1/3 stop. The D3s is inherently 1/3 stop lower in noise. If you agree with my "all else equal" concept that going to DX and dropping the TC is a push in terms of noise, then the D3 D3s (edited) still beats the D7000 by 1/3 stop.
Just above you just got done saying "I am dealing in real world cameras that I have in my possession." but now you immediately thereafter bring up the D400 to muddy the waters. This is what I mean by your moving targets.
We don't know what the D400 will do. We can only assume it will be "all else equal" in noise, square millimeter for square millimeter, compared to the D800 or whatever later FX body may be in hand by the time the D400 rolls out.
>>there is >>one and only one way to reduce noise in the case where >you >>have a fixed size subject and a fixed working distance: >make >>the front element of the lens bigger. > >Disagree here as well. The performance of the lens could also >be increased. That would not show as large a gain as making >the front element larger.
What do you mean by "performance of the lens", if not the light gathering power? The only thing that effects noise is the amount of light and in a long tele that is simply the light gathering power (diameter) of the lens.
>The other issue here, and I touched on it earlier, is that >what you are saying is that pixel density does not have a >bearing on sensor performance for any given generation of >sensor technology. And I believe that to be untrue. Sadly, >there is not an easy, empirical way to test this as >manufacturers tend not to produce the same sized sensor with >varying densities in the same generation of technology. I may >be wrong about this, but if you have strong evidence to the >contrary, I'd love to see it.
Don't tend to produce sensors in the same format with varying densities with the same technology? How about the D4 and D800? Isn't a two month spread in announcement dates close enough? And please do not rehash this thing about longer development cycles for Pro bodies unless you have Nikon documentation indicating the start of the *active* development cycle for those bodies. Your opinion on that is pure speculation based on life cycles, which has no bearing on development or "design lock-in" time tables.
The D4 and D800 are as identical in development era as we could ever see.
Bill Claff rates the D800 as matching the D4, within 0.2 stops of dynamic range from ISO 800 and up. DXO rates the D4 ISO as 2965 vs the D800 ISO of 2853 based on some minimal quality criteria. An ISO diff of 100 at the 2900 level is insignificant.
That is as much proof as anyone could provide that sensor density has no significant fundamental impact on noise. The minor differences in noise performance between those two cameras can easily be accounted for simply because they are different bodies and performance will vary a bit for many reasons.
>My D3s has the best ISO performance of any camera I own. The >D800 has the same sized sensor, and uses exactly the same >glass. It has a sensor that is at least 1, if not two >generations further along. Yet it's ISO performance is not as >good. I am left to conclude that the pixel density plays a >significant role.
You say the D800's performance is "not as good". Bill Claff says it is within 0.2 stops - slightly more than 1/6 stop and probably nothing any of us would ever notice. Only a computer would notice. DXO says the same, per my comment just above.
If you are right, then those two major independent testers are flat out wrong. I have to go with the testers short of substantial evidence from you.
>>All the above is why I previously said that I am >"old >>school" and that I do not believe you can scale up >>sensors indefinitely and transparently trade focal length >for >>pixel density. And why I have trouble with your idea that >the >>Pros are all going to small lenses and DX cameras. > >The pros aren't all going to small lenses and DX cameras. The >pros were forced AWAY from big lenses and DX cameras. To >compensate, they went to even bigger lenses or took a hit in >aperture and put on TCs.
Yes, they "took a hit in aperture" by adding a TC and gained that stop back in the FX advantage in noise performance that allows them to shoot one stop higher ISO, all else equal, but all else was not equal because the D3s totally blows away anything in a DX format.
Nikon actually did them a favor, forcing them to a bigger sensor with more light gathering power, and bigger glass, with more light gathering power, and gave them a system unequaled in image quality.
But no one is ever happy, not even the pros. Not even when they have the best technology man can produce, and at the high end Pro level, the money to fund it.
The first two professional >generations of Nikon's cameras were DX. It wasn't until the >D3 that this changed, and pros were left with the choice of >going with the D3 and losing effective focal length or moving >to the D300 and keeping what they were used to. Of course, at >the expense of ISO performance compared to the larger sensor. >Now that pros have the choice of DX, with the sensor >performance of the former D700/D3 or very nearly so, there is >a lot less pressure to stay full frame.
The D700 has a 2/3 stop advantage over the D7000 so I do not understand this "nearly so". And that is comparing several years later DX technology while FX is also improving in the interim.
> >>You have, in different replies here, suggested FX does >and >>doesn't have excess high ISO performance, depending on >the >>context and shooting situation, which makes it hard for me >to >>keep up with a moving target . But once you show me a >3 >>stop underexposed image as your best way to cope with a >real >>world situation, I gotcha! >
> > >>You can certainly change the FOV via high density DX to >>deceive yourself into thinking you have a "longer >focal >>length" and if your long lens technique is beyond >>reproach you can (in theory, avoiding diffraction issues >as >>you generally can when shooting wide open) transparently >trade >>pixel density for focal length and get equivalent >resolution. > >And this is the game I am playing. Not only this game, but >the game of increased aperture because I get to shoot at F2.8 >instead of F4. > >>There is sort of a free lunch here, but only in regard to >>resolution in terms of pixels across the subject. > >And that... is enough.
No!!!!! That is not enough. You just posted an image you claim is 3 stops underexposed and you are looking to DX to fix it. This is NOT just about pixels across the subject, in fact it has nothing to do with it. it has to do with noise from ridiculously low light levels not even a D3s can deal with it.
I do not believe that image is 3 stops underexposed. If it were the grass would be about luminosity 10 and coal black like the woods in the top half of the scene. I do believe the center weighted meter indicated 3 under but that is only because it averaged in that coal black top half of the scene with the grass and players.
If you had slowed your shutter 3 stops to make the center weight meter happy you would have totally blown out your player's white shirts and the grass would have been at least 1 stop overexposed, probably two. But that is a side issue and not relevant to the problem at hand.
I figure the image at 1.0 to maybe 1.5 at most underexposed. The grass should be luminosity 128 or so, where it is actually about 70. That's about 1 stop. I don't know about you but I use grass as big grey cards.)
>>But consider that in a reply elsewhere here today you >>suggested that FX is near ideal but DX still has some room >to >>improve. If that is the case, then you are also >suggesting >>that the DX/FX trade is even worse than the push I >suggest >>above because the DX sensor inherently does not perform >as >>well as FX on a per square millimeter basis - which is how >the >>"ideal performance" is computed. > >True. And there is a balance point between the advantage of >FOV across a more dense sensor (16MP/24MP DX) versus the less >ideal performance of that chip.
Above you argued repeatedly that you believe denser sensors are noisier. I have argued repeatedly that that is not case, not on a theoretical basis nor based on Bill's empirical tests and DXO's empirical tests of the D4/D800.
However, just for a moment and for arguments sake, let's assume you are right and every authority on the subject is flat out wrong.
Where does that leave you when you take your future brand new 24mpx DX sensor and 300/2.8 onto that dimly lit soccer field? Isn't that new 24 mpx DX sensor going to perform far worse than even I suggest???