Up-Rez and down-rez for comparisons - why?

[dwswager]

If you do not fill your sensor (e.g. BIF) usually the smaller pixel wins, all else equivalent.

Jack

That assumes identical sensor technology and support electronics.  In the real world, I for example have to compare a full frame D810 and the APS-C D7100 and the D810 destroys the D7100 for image quality at higher ISOs. 

[spidermike]

If you do not fill your sensor (e.g. BIF) usually the smaller pixel wins, all else equivalent.

Jack

I'm not so sure. I have seen a few comparisons of the 1Dx, 5D3 and 7D cropped to the same FOV where the noise of the smaller sensor cancels out the greater detail inherent in the higher pixel count. It then comes down to whether you prefer the look of noise or detail.

[Jack Hogan]

That assumes identical sensor technology and support electronics.

Sure, aotbe.

In the real world, I for example have to compare a full frame D810 and the APS-C D7100 and the D810 destroys the D7100 for image quality at higher ISOs.  

Sure about that?  Comparing the same final subject at the same display size?

I'm not so sure. I have seen a few comparisons of the 1Dx, 5D3 and 7D cropped to the same FOV where the noise of the smaller sensor cancels out the greater detail inherent in the higher pixel count. It then comes down to whether you prefer the look of noise or detail.

Re-reading it I realize that my statement wasn't precise enough and could be misinterpreted:

The situation I am referring to is when the desired region of interest does NOT fill the sensor and field of view is NOT the same for the two formats, as is typical in Bird In Flight or distant action captures with the same lens.  Assume the image of the bird on the sensor is the same linear size on both formats in microns (same lens).  The smaller pixel typically wins, because you get more of them on the bird.  The D7100/D7200 are a perfect example, with their 3.9 micron pitch and read noise approaching 1e- away from base ISO.  The 7DI/II comes close in terms of pitch, but at the expense of a noisier pixel.  There is no other Nikon or Canon DSLR that I am aware of that comes close aotbe, for that particular situation.  And sure, I was purposely excluding FPS and AF accuracy from the discussion to keep things simple.

Partly because of that I understand from my friends who do BIF that the D7200 and 7DII are the cameras to have today.  We'll see how quick the 5DS is when it comes out.

Jack

[dwswager]

I'm not so sure. I have seen a few comparisons of the 1Dx, 5D3 and 7D cropped to the same FOV where the noise of the smaller sensor cancels out the greater detail inherent in the higher pixel count. It then comes down to whether you prefer the look of noise or detail.

Most of my sports stuff is destined for screen.  Usually I shoot the D810 is 1.2X crop mode for sports.  That gives basically a 1.2X crop 24MP image while the D7100 gives a  1.5x crop 24MP image.  The D810 pixels are 4.8µm while the D7100 pixels are 3.9µm. So I usually get more pixels on target with the D7100, but the quality of the pixels is not as good.  The comparison might be different if I was shooting a 300 or 400mm f/2.8 where I could get plenty pixels on target and down sample significantly.

It interesting that I typically see a lot of 7Ds out during the day, but at night sports, it is almost exclusively FF cameras.  Some 1Dx and D4s, D3x and D810 and D750s.  Of the 4 guys shooting 7D in the day only one shoots at night and he brings a 1Dx.  I would love a DX camera that controlled noise somewhere near where the D810.

[Iluvmycam]

OP....I don't know why. I just compare them, I don't doctor them.

http://photographycompared.tumblr.com/

[Jack Hogan]

...
The D810 pixels are 4.8µm while the D7100 pixels are 3.9µm. So I usually get more pixels on target with the D7100, but the quality of the pixels is not as good.  The comparison might be different if I was shooting a 300 or 400mm f/2.8 where I could get plenty pixels on target and down sample significantly.
...
I would love a DX camera that controlled noise somewhere near where the D810.

I hear you in general, but do we agree that a 24MP APS-C sensor with similar per pixel metrics - with the same lens at the same focal length, shutter speed and f-stop - would show better resolution and just as well controlled noise in a situation such as BIF, when the D810 is forced to shed its larger sensor advantage because the image of the region of interest projected onto the sensing plane fits comfortably inside the 15.6x23.5mm DX sensor area (corresponding to about 3200x4800= 15.4MP on the D810), so one would want to crop it to that resolution anyways?  And aren't D7200 per pixel metrics actually better than the D810's throughout the ISO range, the 7DII's catching and surpassing it by ISO1600 or so?  Again ignoring FPS, AF and other issues for argument's sake?

I assume you know all this but I am asking because many people never realize this simple fact, and as a result sometimes end up using the wrong tool for the job.

Jack

[dwswager]

I hear you in general, but do we agree that a 24MP APS-C sensor with similar per pixel metrics - with the same lens at the same focal length, shutter speed and f-stop - would show better resolution and just as well controlled noise in a situation such as BIF, when the D810 is forced to shed its larger sensor advantage because the image of the region of interest projected onto the sensing plane fits comfortably inside the 15.6x23.5mm DX sensor area (corresponding to about 3200x4800= 15.4MP on the D810), so one would want to crop it to that resolution anyways?  And aren't D7200 per pixel metrics actually better than the D810's throughout the ISO range, the 7DII's catching and surpassing it by ISO1600 or so?  Again ignoring FPS, AF and other issues for argument's sake?

I assume you know all this but I am asking because many people never realize this simple fact, and as a result sometimes end up using the wrong tool for the job.

Jack

Absolutely, in the case of the D7100 versus D810, the D7100 is giving more pixels of smaller size on target leading to more detail while the larger pixel size of the D810 is giving lower noise.  But this is not constant over all shots.  The DX camera reach makes some shots possible that I just don't pull the trigger on the FF if it is so small in the frame and close up, I get more pixels on target of larger size with the full frame.  The action is never a constant distance away.  Hence, I favor the actual noise advantage of the newer, better sensor with larger pixel size at night.  In the daytime, I usually go with the D7100 where I get the magnification.  Though I have to say the D810 is actually faster operating which I attribute to the better processor and larger buffer.

BTW, for a FF and DX sensored cameras to have identical pixel size, then the full frame would need to be approximately 48MP versus 24MP for the DX (1.5X crop) camera.  In that situation, both cameras would be putting the same number of same size pixels on target and neither camera (theoretically) should have a noise advantage considering similar size pixels of the same technologies.

[Jack Hogan]

Absolutely, in the case of the D7100 versus D810, the D7100 is giving more pixels of smaller size on target leading to more detail while the larger pixel size of the D810 is giving lower noise.

This is the point I was trying to make: the bigger pixels of the  D810 are no help in the situation described, in fact its higher relative read noise puts it at a disadvantage where noise performance is concerned.

The desired sensing area is the same, therefore so is the total number of photons counted in that area (because we have said f/stop, shutter speed and all other things are the same).  So when the resulting BIF image is shown with the same number of pixels at the same size the better overall noise performance goes to the camera with the better effective read noise per area, independently of actual pixel size.  In this case that's the D7200 (or the D7100, but let's not bring banding into the discussion).  In this particular situation, typically only found in BIF and distant action, the DX camera with the same glass is better, period.  In noise and resolution.  Throughout the working ISO range (say up to 6400). During the day or at night.  The 1DX need not apply (well, ok, but at higher ISOs only if you do not mind the resolution hit).

BTW, for a FF and DX sensored cameras to have identical pixel size, then the full frame would need to be approximately 48MP versus 24MP for the DX (1.5X crop) camera.  In that situation, both cameras would be putting the same number of same size pixels on target and neither camera (theoretically) should have a noise advantage considering similar size pixels of the same technologies.

Right.

Jack

PS  A few numbers extracted from DxOmark.com to back up the rhetoric.

[dwswager]

PS  A few numbers extracted from DxOmark.com to back up the rhetoric.

Yes, if I was shooting a D7200 that uses a newer Sony Exmor, that would be correct, but the 2 year old D7100 that I have uses a Toshiba Sensor that does not perform as well as the D810.

However, the good point in your post is that we are getting better sensors all the time.  Sony continues to improve to the point that even at a smaller pixel size, the  D7200 APS-C is able to out perform a modestly older sensor with bigger pixels.  That is fabulous for photographers and should be the impetus necessary for Nikon to release a professional level DX body!

[Jack Hogan]

Yes, if I was shooting a D7200 that uses a newer Sony Exmor, that would be correct, but the 2 year old D7100 that I have uses a Toshiba Sensor that does not perform as well as the D810.

The D7100 data I have, elaborated from captures by Bill Claff, is so similar to the D7200 I would be surprised if it weren't a tweaked version of the same sensor with slightly lower read noise and banding fixed.  Happy to learn otherwise if that's not the case.  Here is an example at base ISO obtained with a different methodology than the earlier ones:



You can find figures for the other ISOs at sensorgen.info, obtained from DxO data.  Ignoring banding for the sake of the argument, its read noise is better on a per area basis than the D810's.  As you can see the idea that the newer sensor has lower per pixel read noise independently of size is incorrect: statistically (if not for technical reasons) at working ISOs read noise has scaled somewhat with pixel size for years. Look at the EM-1 for another example, which I believe wears Sony and was introduced the year before the D810.

Anyways, I believe we understand each other and the point: the D810 is arguably the best general purpose camera on the market today.  However, for a distant subject that only fills the DX field of view one can do better in terms of noise and resolution, aotbe.

Jack

[dwswager]

The D7100 data I have, elaborated from captures by Bill Claff, is so similar to the D7200 I would be surprised if it weren't a tweaked version of the same sensor with slightly lower read noise and banding fixed.  Happy to learn otherwise if that's not the case.  Here is an example at base ISO obtained with a different methodology than the earlier ones:



You can find figures for the other ISOs at sensorgen.info, obtained from DxO data.  Ignoring banding for the sake of the argument, its read noise is better on a per area basis than the D810's.  As you can see the idea that the newer sensor has lower per pixel read noise independently of size is incorrect: statistically (if not for technical reasons) at working ISOs read noise has scaled somewhat with pixel size for years. Look at the EM-1 for another example, which I believe wears Sony and was introduced the year before the D810.

Anyways, I believe we understand each other and the point: the D810 is arguably the best general purpose camera on the market today.  However, for a distant subject that only fills the DX field of view one can do better in terms of noise and resolution, aotbe.

Jack

Thanks!  Maybe I'm just shooting the D7100 wrong!  I will have to go out and retest it.  It would be great to have the extra reach the DX crop gives versus the 1.2x crop I get from the D810.  I will say, the D810 operates faster and has a much better buffer.  Former D300 shooter still waiting for the update! Doh!

[dwswager]

The D7100 data I have, elaborated from captures by Bill Claff, is so similar to the D7200 I would be surprised if it weren't a tweaked version of the same sensor with slightly lower read noise and banding fixed.  Happy to learn otherwise if that's not the case.  Here is an example at base ISO obtained with a different methodology than the earlier ones:




Jack

I do keep coming back to the DxOMark SNR data showing the D810 about 1.3 stops better than either the D7100 or 7DmkII.  This is what I see in photos between my D810 and D7100.  Typically, the D810 performs (For Me as I am shooting them) at about 1.5 - 2 stops better than the D7100.

[ErikKaffehr]

Hi,

SNR is essentially proportional to sensor size. If you compare at actual pixels the results will be somewhat different. The Nikon still leads at it has larger pixels than the Canon D7II or the D7200. When you shoot the Nikon in crop mode you loose the advantage of size.

I would also say that the sensor in the Nikon D810 is probably the best one Sony ever made. The 24 MP sensors seem to be behind in performance.

Best regards
Erik

I do keep coming back to the DxOMark SNR data showing the D810 about 1.3 stops better than either the D7100 or 7DmkII.  This is what I see in photos between my D810 and D7100.  Typically, the D810 performs (For Me as I am shooting them) at about 1.5 - 2 stops better than the D7100.

[Jack Hogan]

I do keep coming back to the DxOMark SNR data showing the D810 about 1.3 stops better than either the D7100 or 7DmkII.  This is what I see in photos between my D810 and D7100.  Typically, the D810 performs (For Me as I am shooting them) at about 1.5 - 2 stops better than the D7100.

Yes, if your subject fills the FF field of view, that is what you will get.  But in BIF or distant subjects that only fill the APS-C field of view this is what you get instead:



And this is before accounting for the fact that in this situation a D7x00 would produce 1.5x as many pixels as a D8x0.

Jack

[Jack Hogan]

I would also say that the sensor in the Nikon D810 is probably the best one Sony ever made. The 24 MP sensors seem to be behind in performance.

Can you elaborate on that Erik?

Jack

[ErikKaffehr]

Hi Jack,

Thanks for asking…

Well, what I have seen is that the 36MP sensors seem to be at the top of the DxO-mark ratings and I would say that I have the impression that the Sony A7r is ahead the normal A7 in most respects. Also, 36 MP is quite a bit more than 24 MP, if you need it, that is.

So, what I see is that Sony has a 36 MP sensor a couple years old that still outperforms the 24 MP-sensors. Now, that depends on viewpoint, of course. I am a tripod shooter so I have little interest in high ISO performance for instance.

So, it is an impression and it is based on my preferences. I would also add that I don't own any of those 36 MP cameras. I may jump on that wagon train, once Sony builds a body I want to buy. Right now I have my Hasselblad with my 39 MP back. I wouldn't switch to Canon or Nikon, DSLRs is not where I think the future is.

Best regards
Erik

Can you elaborate on that Erik?

Jack

[Jack Hogan]

So, what I see is that Sony has a 36 MP sensor a couple years old that still outperforms the 24 MP-sensors. Now, that depends on viewpoint, of course. I am a tripod shooter so I have little interest in high ISO performance for instance.

I understand, I am a tripod shooter too.  I asked because my recollection was that for all intents and purposes with regards to noise the size of the pixel seemed to make very little difference when looking at images displayed at the same size, all other things (e.g. format, manufacturer and technology generation) equal.  So I went back and looked at DxO data:



The only surprise there (at least to me) was the fact that the D810's could reach a lower base ISO.  SNR curves (not shown) pretty well overlay each other, except that the D810's keeps going the extra little bit, as above.  So what's going on?  It may be informative to look at some key parameters of current FF Sony sensors on a per unit area basis, data from sensorgen.info at ISO 100:



By and large the same performance: at ISO 100 all sensors shown saturate more or less at the same exposure and show roughly the same read noise and clipping in units of photoelectrons per micron squared - with perhaps the D810 a little lower than the average in both RN and FWC.  DRs per unit area are directly comparable and very similar, within the margin of error.  Looking at this ISO 100 chart, and ignoring resolution, there appears to be little or no difference in performance whether pixels of the Sony FF sensors of the last couple of generations are larger or smaller, as predicted by DxO's chart.  But let's take a closer look at the D810 and its extended base ISO, which is substantially lower than the others'.

D810 @ISO64: Hsat 1.66 lx-s, RN 2.49 e-/um^2, clipping 3296 e-/um^2, DR 10.4 stops/um^2

Notice the two outlying parameters in bold above, about 50% better than their peers?  They indicate a change in technology.  There are reports that Sony had a patent swap with Aptina a couple of years ago, and the technology necessary for this type of base ISO extension is described in one well known Aptina patent.  Incidentally, it also appears that the A7s uses a similar trick, albeit to improve high ISO performance as opposed to extend base.

So do you see why I thought that captures of sensors of the same generation using the same technology result pretty well in the same noise performance when displayed at the same size?  The D810 is an outlier because it uses newer technology than the others.  Hopefully it will trickle down to all other sensors by the same manufacturer as well in the future.  Keep this (and shutter shock) in mind when making your choices.

Cheers,
Jack

[ErikKaffehr]

Hi Jack,

Thanks for the detailed analysis. Some reflections:

• I have also noticed that SNR curves for same size sensor are pretty similar, so I assumed that development was mostly on lowering readout noise
• I have also noticed that the newer sensors have lower base ISO.

Base ISO (according to DxO)

Nikon D810 - 47
Sony  A7r   - 73
Leica M240 - 134

The SNR curve is essentially the same, indicating similar quantum efficiency, I think. But Nikon jumps earlier on the train than Sony does and Leica is a late arriver. I would think the development is increased well capacity.

Best regards
Erik

I understand, I am a tripod shooter too.  I asked because my recollection was that for all intents and purposes with regards to noise the size of the pixel seemed to make very little difference when looking at images displayed at the same size, all other things (e.g. format, manufacturer and technology generation) equal.  So I went back and looked at DxO data:



The only surprise there (at least to me) was the fact that the D810's could reach a lower base ISO.  SNR curves (not shown) pretty well overlay each other, except that the D810's keeps going the extra little bit, as above.  So what's going on?  It may be informative to look at some key parameters of current FF Sony sensors on a per unit area basis, data from sensorgen.info at ISO 100:



By and large the same performance: at ISO 100 all sensors shown saturate more or less at the same exposure and show roughly the same read noise and clipping in units of photoelectrons per micron squared - with perhaps the D810 a little lower than the average in both RN and FWC.  DRs per unit area are directly comparable and very similar, within the margin of error.  Looking at this ISO 100 chart, and ignoring resolution, there appears to be little or no difference in performance whether pixels of the Sony FF sensors of the last couple of generations are larger or smaller, as predicted by DxO's chart.  But let's take a closer look at the D810 and its extended base ISO, which is substantially lower than the others'.

D810 @ISO64: Hsat 1.66 lx-s, RN 2.49 e-/um^2, clipping 3296 e-/um^2, DR 10.4 stops/um^2

Notice the two outlying parameters in bold above, about 50% better than their peers?  They indicate a change in technology.  There are reports that Sony had a patent swap with Aptina a couple of years ago, and the technology necessary for this type of base ISO extension is described in one well known Aptina patent.  Incidentally, it also appears that the A7s uses a similar trick, albeit to improve high ISO performance as opposed to extend base.

So do you see why I thought that captures of sensors of the same generation using the same technology result pretty well in the same noise performance when displayed at the same size?  The D810 is an outlier because it uses newer technology than the others.  Hopefully it will trickle down to all other sensors by the same manufacturer as well in the future.  Keep this (and shutter shock) in mind when making your choices.

Cheers,
Jack