Dual-pixel AF and ISO performance?

[Loren E]

I switched from Nikon to Canon recently, and for my crop body went from a D7100 to a 7D. The 7D seems great with the major exception of high ISO performance, which I find disappointing compared to the D7100. Yes of course the D7100 is much newer sensor technology, so it should be better at high ISO, but I have high hopes of a 7d mk II in 2014 that would trump the D7100 at high ISO because of even newer sensor tech.

But now with the 70D having dual pixel AF, which is of little use to me since I don't shoot video, I assume this will be in the 7d mk II as well. My question is: Does dual pixel AF have the downside of hindering high ISO performance? I have read this from uninformed sources and am curious if there is truth to it. As a still shooter, I would be way bummed if technology that mostly benefits video shooters was really hurting us still shooters! I want the best high ISO performance possible in a crop body for stills and would be thrilled to see a 2 stop improvement over the current 7D, but man what a bummer if we saw only minor improvement as a result of dual-pixel tech implementation. (would be WAY cool if Canon offered 2 versions, a video optimized version and a sills optimized version). And before you tell me to just go full frame if I want quality high ISO, I own a 5d mk III and love its high ISO, but will always switch to the crop when I am shooting wildlife or other reach-limited scenarios.

Thanks for any clarification on this dual-pixel AF tech question!

 

[Mt Spokane Photography]

The Dual pixel does not hinder AF performance, but if high ISO interests you, just go to FF.

 

[xvnm]

It is very hard, if not impossible, to say what kind of ISO performance the 70D would have if it didn't have DPAF.

With DPAF, the 70D has better ISO performance than the 7D/60D, although the difference is minor, almost academic.

It's reasonable to assume the 7D II will have better ISO than the 7D, even if slight. Whether it will be 2 stop worth, your guess is as good as mine.

 

[candc]

I have a 70d and the iso performance is maybe 1 stop better than 7d? If the 7dii is one better yet then that would be fantastic. I would trade mp for iso performance. I would be happy with a 10mp camera like the 40d with the same photosites on the 6d. The 40d has all the resolution you need if you are not heavy cropping and you would have really good low light performance to go with it. They won't go that extreme but hopefully they will keep the mp the same or slightly lower as they did in the change from 1dsiii to 1dx

 

[Loren E]

The Dual pixel does not hinder AF performance, but if high ISO interests you, just go to FF.

Heh don't think you read much of the post. Especially this part " And before you tell me to just go full frame if I want quality high ISO, I own a 5d mk III and love its high ISO, but will always switch to the crop when I am shooting wildlife or other reach-limited scenarios. "


From the other posters it sounds like it isn't known to the public whether Dual-pixel AF hinders ISO performance capabilities, but can't be a good sign if the 70D is that little of an improvement unless they independently crippled its high ISO performance to help carve out a niche for the 7d mk II.

 

[Lawliet]

But now with the 70D having dual pixel AF, which is of little use to me since I don't shoot video, I assume this will be in the 7d mk II as well. My question is: Does dual pixel AF have the downside of hindering high ISO performance?

Noise consist of three components - photon noise can't be helped, you need more photons to get rid of it; noise in the sensels, where dpAF makes things worse but where Canons sensors are quite good to start with; and readout noise where Canons achilles heal currently lies but where one can use the split readout to average the noise out. Not much to worry about. OTOH: a subgroup of still photographers gets more accurate AF, making better use of the resolution those expensive lenses offer...if focussing is spot on. Throw more processing power at the problem and the action orientated folks might see the same benefit.

 

[jrista]

It is very hard, if not impossible, to say what kind of ISO performance the 70D would have if it didn't have DPAF.

With DPAF, the 70D has better ISO performance than the 7D/60D, although the difference is minor, almost academic.

It's reasonable to assume the 7D II will have better ISO than the 7D, even if slight. Whether it will be 2 stop worth, your guess is as good as mine.

I don't think it is physically possible for any APS-C camera with the same or smaller pixel pitch to have a true two stop high ISO advantage. By that, I mean that ISO 3200 on the 7D II could never be one quarter as noisy as the 7D classic. The amount of noise at high ISO is dominated by photon shot noise, which derives from the physical nature of light itself...it therefor has little to do with the sensor technology with one exception.

The 7D has a quantum efficiency (Q.E., the efficacy to convert photons to electrical charge) of 41%. That means, for the photons actually reaching the photodiode, 41% of them on average are converted into charge. To truly "double" (improve by one stop) the high ISO performance of the 7D II, Q.E. would have to be 82%. To double that, Q.E. would have to be 164%...something that plain and simply isn't possible. At best, you could concert 100% of the photons in an ideal model, but in reality you are probably limited to 99%. That gets you about 20% of the next stop improvement. So, at the absolute best, assuming Canon achieved one hell of an amazing feat, the 7D II could have ISO 3200 performance that was 1 1/5th stops better than the 7DC.

The best, scientific grade CCDs are supercooled down to -80°C or so, which nearly eliminates dark current noise (by around 200x less dark current a room temperature sensor has at 72°C). These CCDs have around 80-85% Q.E. The equipment required to cool a sensor that much is large, bulky, and draws an immense amount of power. Given that, it is highly doubtful we'll see a true "one stop" improvement in high ISO performance. We might see two additional stops of selectable high ISO settings, up to 25,600...however those stops won't perform much better than they do on any other camera. Canon may get Q.E. up into the 50-60% range, which might give you a fifth to a quarter of a stop noise improvement...very small, but probably slightly noticeable.

This improvement shouldn't have anything to do with whether DPAF is used or not. DPAF might reduce Q.E. by a fraction, a percent, depending on exactly how it is designed and whether it causes a loss in photon conversions. Regardless, I wouldn't expect the 7D II to get any kind of truly significant improvement in high ISO performance unless its megapixel count drops and it's pixels get bigger (as increasing pixel area is the only surefire way to improve the amount of light gathered per pixel regardless of what your Q.E. is...which is why larger sensors with bigger pixels tend to do better at high ISO.)

 

[Mt Spokane Photography]

The Dual pixel does not hinder AF performance, but if high ISO interests you, just go to FF.

Heh don't think you read much of the post. Especially this part " And before you tell me to just go full frame if I want quality high ISO, I own a 5d mk III and love its high ISO, but will always switch to the crop when I am shooting wildlife or other reach-limited scenarios. "


From the other posters it sounds like it isn't known to the public whether Dual-pixel AF hinders ISO performance capabilities, but can't be a good sign if the 70D is that little of an improvement unless they independently crippled its high ISO performance to help carve out a niche for the 7d mk II.

Let me repeat it, Dual Pixel does not hinder the ISO. Its been well tested, and results are easy to find online. It also does not improve ISO, but you did not ask that!

Switching from a 5D MK III to say a 7D in bright light is fine for focal length limited scenarios where you can use low ISO. However, at High ISO, you will gain due to lower noise. You can also AF to f/8 with your 5D MK III which may be a benefit.

 

[jrista]

I have a 70d and the iso performance is maybe 1 stop better than 7d? If the 7dii is one better yet then that would be fantastic. I would trade mp for iso performance. I would be happy with a 10mp camera like the 40d with the same photosites on the 6d. The 40d has all the resolution you need if you are not heavy cropping and you would have really good low light performance to go with it. They won't go that extreme but hopefully they will keep the mp the same or slightly lower as they did in the change from 1dsiii to 1dx

why would trade anything for resolution?
it is a myth that lower resolution gains higher iso

I wouldn't say that. With smaller pixels you have more area where light can be lost. More r/c activate and readout wiring occupies space that was once just pure pixel area. Pixel wells effectively become deeper (in relative terms), limiting the angle of incidence that can be successfully captured. Microlensing helps, but it still isn't as good as having purely larger pixel area. Lightpipes can reduce losses, but still, not as good as raw pixel area. In that respect alone, smaller pixels for any given format are never going to experience a purely linear drop in per-pixel noise...it tends to be a little higher.

This can be demonstrated with any two cameras if you know some details about their sensor design. If we pick the 5D III and 7D, for example. The 5D III pixel pitch is 6.25µm, the 7D pixel pitch is 4.16µm. The 5D III pixel saturation point in electrons is 67531e-, the 7D pixel saturation point is 20187e-. The pixel area ratio between the two is 2.26x, however the saturation ratio is 3.35x. There is a slight difference in Q.E., with the edge going to the 5D III, but it isn't significant enough to produce the skew demonstrated in the pixels charge capacity.

There isn't a linear scale in pixel saturation, which one would have to assume if "it is a myth that lower resolution gains higher ISO." A linear scaling would be necessary to ensure that downsampling a higher resolution sensor image to the same output dimensions of a lower resolution sensor resulted in the same noise.

Theoretically, "all else being equal", then yes, you could always downsample a higher resolution image to the same dimensions as a lower resolution, and basic averaging should result in similar noise. Simple fact is that as pixels scale down, pixel area scales down faster, until you finally reach a point where you have to take drastic measures to produce viable pixel area...say using a BSI design. Since BSI has not yet been used in APS-C and FF cameras, there isn't any real empirical data to use to determine if a BSI sensor would experience enough of a performance boost to maintain a more linear scaling of ISO performance.

 

[Loren E]

You said "The Dual pixel does not hinder AF performance" and then said "Let me repeat it, Dual Pixel does not hinder the ISO"....did my confusion arise because you meant to say ISO performance instead of AF performance in your first post?

And hasn't the Nikon D7100 been tested to be close to 2 stops better at high ISO than the 7d? I didn't do any side by side comparisons but definitely felt like my level of satisfaction with ISO 6400 on the D7100 was similar to that felt with ISO 1600 on the 7D...at least in terms of where I felt I wanted to cap my high ISO usage for the respective bodies....maybe to be fair 1600 on the 7D is more like 4000 on the D7100 than 6400.

The Dual pixel does not hinder AF performance, but if high ISO interests you, just go to FF.

Heh don't think you read much of the post. Especially this part " And before you tell me to just go full frame if I want quality high ISO, I own a 5d mk III and love its high ISO, but will always switch to the crop when I am shooting wildlife or other reach-limited scenarios. "


From the other posters it sounds like it isn't known to the public whether Dual-pixel AF hinders ISO performance capabilities, but can't be a good sign if the 70D is that little of an improvement unless they independently crippled its high ISO performance to help carve out a niche for the 7d mk II.

Let me repeat it, Dual Pixel does not hinder the ISO. Its been well tested, and results are easy to find online. It also does not improve ISO, but you did not ask that!

Switching from a 5D MK III to say a 7D in bright light is fine for focal length limited scenarios where you can use low ISO. However, at High ISO, you will gain due to lower noise. You can also AF to f/8 with your 5D MK III which may be a benefit.

 

[jrista]

You said "The Dual pixel does not hinder AF performance" and then said "Let me repeat it, Dual Pixel does not hinder the ISO"....did my confusion arise because you meant to say ISO performance instead of AF performance in your first post?

And hasn't the Nikon D7100 been tested to be close to 2 stops better at high ISO than the 7d? I didn't do any side by side comparisons but definitely felt like my level of satisfaction with ISO 6400 on the D7100 was similar to that felt with ISO 1600 on the 7D...at least in terms of where I felt I wanted to cap my high ISO usage for the respective bodies....maybe to be fair 1600 on the 7D is more like 4000 on the D7100 than 6400.

There is perceptually better and actually better. There is most definitely a psychological component to thinking that the D7100 is "two stops" better at high ISO. From a technical standpoint, it probably isn't possible to actually get truly two stops better, since stops are a power of two, and ISO performance is dependent upon Q.E. and pixel area. The 7D has larger pixels (4.16µm vs. 3.91µm), which is it's benefit, where as the D7100 has more Q.E. (but certainly not enough to literally be two stops better.) The D7100 has 11% better Q.E. than the 7D (52% vs. 41%). In terms of pixel area, the 7D pixels are 13% larger. Technologically, the D7100 has a better sensor with a better architecture, which also probably gives it an edge when it comes to high ISO (primarily, it has a higher SNR, which means that at every ISO, it has a larger usable signal). Overall, from a literal, physical, technical standpoint, the difference between these two sensors is fairly small, and while the D7100, thanks to its excellent SNR, does better, it isn't anywhere close to two stops (i.e. the 7D at ISO 6400 has a saturation point of 536 vs. the D7100 at ISO 6400 which has a saturation point of 541...almost negligible.)

To truly have a full two stops better noise performance, where the amount of noise at ISO 6400 is the same as the amount of noise at ISO 1600, you either need to reduce megapixel count by a factor of two (pixels that are four times greater area...i.e. a pixel pitch of 7.82µm)...or you need to increase quantum efficiency by two orders of magnitude. The 7D has a Q.E. of 41%. Twice the efficiency is 82%. Twice that is 164%. Well, it's impossible to gather more photons than exist, so you can't have more than 100% Q.E. (and achieving that usually requires rather bulky cooling equipment that would render such a camera immobile.)

From a perceptual standpoint...softer detail appears to suffer more from noise. The 7D has an AA filter, where as the D7100 does not. The D7100 is going to have much sharper detail due to having more acutance. THIS is its true strong point when it comes to ISO performance, and probably the key reason why it "feels" as though it has less noise. Detail is sharper with the D7100, so noise doesn't appear to be as prevalent, even though it is roughly the same as the 7D. There is a tradeoff for this...more aliasing and moire. General aliasing can be delt with to a degree with downsampling, moire can be very difficult to deal with (there are some tools, however most simply reduce color moire and mitigate monochrome moire, but none can actually eliminate it.)

If you don't shoot subjects that have repeating patterns or clean edges that might result in aliasing, then the D7100 is certainly an amazing camera, and its sharper detail will certainly result in perceptually less noise.

The 70D is still using the same general sensor design and architecture as the 7D, so it is doubtful much of its weaknesses have been resolved. I get the feeling that the 70D is sharper, which will go a long way to mitigating how noisy it "feels". It has a larger signal, however again thanks to Canon's read-noisy archaic sensor+ADC architecture, it still isn't as good as the D7100 (26726e- vs. 29236e- FWC.) The 70D also still uses an AA filter, which is going to soften detail around nyquist...and while that eliminates (or greatly reduces) aliasing, the lower acutance will still make it "feel" as though it is noisier...dual pixel architecture or not.

 

[Loren E]

You said "The Dual pixel does not hinder AF performance" and then said "Let me repeat it, Dual Pixel does not hinder the ISO"....did my confusion arise because you meant to say ISO performance instead of AF performance in your first post?

And hasn't the Nikon D7100 been tested to be close to 2 stops better at high ISO than the 7d? I didn't do any side by side comparisons but definitely felt like my level of satisfaction with ISO 6400 on the D7100 was similar to that felt with ISO 1600 on the 7D...at least in terms of where I felt I wanted to cap my high ISO usage for the respective bodies....maybe to be fair 1600 on the 7D is more like 4000 on the D7100 than 6400.

There is perceptually better and actually better. There is most definitely a psychological component to thinking that the D7100 is "two stops" better at high ISO. From a technical standpoint, it probably isn't possible to actually get truly two stops better, since stops are a power of two, and ISO performance is dependent upon Q.E. and pixel area. The 7D has larger pixels (4.16µm vs. 3.91µm), which is it's benefit, where as the D7100 has more Q.E. (but certainly not enough to literally be two stops better.) The D7100 has 11% better Q.E. than the 7D (52% vs. 41%). In terms of pixel area, the 7D pixels are 13% larger. Technologically, the D7100 has a better sensor with a better architecture, which also probably gives it an edge when it comes to high ISO (primarily, it has a higher SNR, which means that at every ISO, it has a larger usable signal). Overall, from a literal, physical, technical standpoint, the difference between these two sensors is fairly small, and while the D7100, thanks to its excellent SNR, does better, it isn't anywhere close to two stops (i.e. the 7D at ISO 6400 has a saturation point of 536 vs. the D7100 at ISO 6400 which has a saturation point of 541...almost negligible.)

To truly have a full two stops better noise performance, where the amount of noise at ISO 6400 is the same as the amount of noise at ISO 1600, you either need to reduce megapixel count by a factor of two (pixels that are four times greater area...i.e. a pixel pitch of 7.82µm)...or you need to increase quantum efficiency by two orders of magnitude. The 7D has a Q.E. of 41%. Twice the efficiency is 82%. Twice that is 164%. Well, it's impossible to gather more photons than exist, so you can't have more than 100% Q.E. (and achieving that usually requires rather bulky cooling equipment that would render such a camera immobile.)

From a perceptual standpoint...softer detail appears to suffer more from noise. The 7D has an AA filter, where as the D7100 does not. The D7100 is going to have much sharper detail due to having more acutance. THIS is its true strong point when it comes to ISO performance, and probably the key reason why it "feels" as though it has less noise. Detail is sharper with the D7100, so noise doesn't appear to be as prevalent, even though it is roughly the same as the 7D. There is a tradeoff for this...more aliasing and moire. General aliasing can be delt with to a degree with downsampling, moire can be very difficult to deal with (there are some tools, however most simply reduce color moire and mitigate monochrome moire, but none can actually eliminate it.)

If you don't shoot subjects that have repeating patterns or clean edges that might result in aliasing, then the D7100 is certainly an amazing camera, and its sharper detail will certainly result in perceptually less noise.

The 70D is still using the same general sensor design and architecture as the 7D, so it is doubtful much of its weaknesses have been resolved. I get the feeling that the 70D is sharper, which will go a long way to mitigating how noisy it "feels". It has a larger signal, however again thanks to Canon's read-noisy archaic sensor+ADC architecture, it still isn't as good as the D7100 (26726e- vs. 29236e- FWC.) The 70D also still uses an AA filter, which is going to soften detail around nyquist...and while that eliminates (or greatly reduces) aliasing, the lower acutance will still make it "feel" as though it is noisier...dual pixel architecture or not.

Really informative post, very interesting about the affect of an AA filter on noise perception. I wonder if Canon will begin going the Nikon route and moving AA filters.

 

[candc]

The dual pixel architecure does not seem to hurt iso performance. The 70d is better than its predecessors. It seems to me that in a given generation of cameras that iso performance is most Dependant on pixel density. DPR has a good tool for doing side by side comparisons This is d800, d7100, 70d, and 5d3 at iso 6400. The first 3 have about the same pixel pitch and look about the same, the 5d3 has the lowest pixel density and looks the best.

http://www.dpreview.com/reviews/image-comparison?attr18=daylight&attr13_0=nikon_d800&attr13_1=nikon_d7100&attr13_2=canon_eos70d&attr13_3=canon_eos5dmkiii&attr15_0=raw&attr15_1=raw&attr15_2=raw&attr15_3=raw&attr16_0=6400&attr16_1=6400&attr16_2=6400&attr16_3=6400&normalization=full&widget=1&x=0.10610376873861654&y=-0.9681097645669836

 

[jrista]

You said "The Dual pixel does not hinder AF performance" and then said "Let me repeat it, Dual Pixel does not hinder the ISO"....did my confusion arise because you meant to say ISO performance instead of AF performance in your first post?

And hasn't the Nikon D7100 been tested to be close to 2 stops better at high ISO than the 7d? I didn't do any side by side comparisons but definitely felt like my level of satisfaction with ISO 6400 on the D7100 was similar to that felt with ISO 1600 on the 7D...at least in terms of where I felt I wanted to cap my high ISO usage for the respective bodies....maybe to be fair 1600 on the 7D is more like 4000 on the D7100 than 6400.

There is perceptually better and actually better. There is most definitely a psychological component to thinking that the D7100 is "two stops" better at high ISO. From a technical standpoint, it probably isn't possible to actually get truly two stops better, since stops are a power of two, and ISO performance is dependent upon Q.E. and pixel area. The 7D has larger pixels (4.16µm vs. 3.91µm), which is it's benefit, where as the D7100 has more Q.E. (but certainly not enough to literally be two stops better.) The D7100 has 11% better Q.E. than the 7D (52% vs. 41%). In terms of pixel area, the 7D pixels are 13% larger. Technologically, the D7100 has a better sensor with a better architecture, which also probably gives it an edge when it comes to high ISO (primarily, it has a higher SNR, which means that at every ISO, it has a larger usable signal). Overall, from a literal, physical, technical standpoint, the difference between these two sensors is fairly small, and while the D7100, thanks to its excellent SNR, does better, it isn't anywhere close to two stops (i.e. the 7D at ISO 6400 has a saturation point of 536 vs. the D7100 at ISO 6400 which has a saturation point of 541...almost negligible.)

To truly have a full two stops better noise performance, where the amount of noise at ISO 6400 is the same as the amount of noise at ISO 1600, you either need to reduce megapixel count by a factor of two (pixels that are four times greater area...i.e. a pixel pitch of 7.82µm)...or you need to increase quantum efficiency by two orders of magnitude. The 7D has a Q.E. of 41%. Twice the efficiency is 82%. Twice that is 164%. Well, it's impossible to gather more photons than exist, so you can't have more than 100% Q.E. (and achieving that usually requires rather bulky cooling equipment that would render such a camera immobile.)

From a perceptual standpoint...softer detail appears to suffer more from noise. The 7D has an AA filter, where as the D7100 does not. The D7100 is going to have much sharper detail due to having more acutance. THIS is its true strong point when it comes to ISO performance, and probably the key reason why it "feels" as though it has less noise. Detail is sharper with the D7100, so noise doesn't appear to be as prevalent, even though it is roughly the same as the 7D. There is a tradeoff for this...more aliasing and moire. General aliasing can be delt with to a degree with downsampling, moire can be very difficult to deal with (there are some tools, however most simply reduce color moire and mitigate monochrome moire, but none can actually eliminate it.)

If you don't shoot subjects that have repeating patterns or clean edges that might result in aliasing, then the D7100 is certainly an amazing camera, and its sharper detail will certainly result in perceptually less noise.

The 70D is still using the same general sensor design and architecture as the 7D, so it is doubtful much of its weaknesses have been resolved. I get the feeling that the 70D is sharper, which will go a long way to mitigating how noisy it "feels". It has a larger signal, however again thanks to Canon's read-noisy archaic sensor+ADC architecture, it still isn't as good as the D7100 (26726e- vs. 29236e- FWC.) The 70D also still uses an AA filter, which is going to soften detail around nyquist...and while that eliminates (or greatly reduces) aliasing, the lower acutance will still make it "feel" as though it is noisier...dual pixel architecture or not.

Really informative post, very interesting about the affect of an AA filter on noise perception. I wonder if Canon will begin going the Nikon route and moving AA filters.

Honestly, I hope they don't remove them. Maybe slightly weaken them, but aliasing and moire are really no better than noise, and in some respects worse...they are harder to deal with in post. Noise is pretty easy to remove, and a moderate amount of removal can greatly reduce noise without hurting image detail. There are no real effective ways to fix aliased edges or remove moire. An AA filter is the only real way to ensure that around nyquist, what the sensor resolves is "realistic". I think Canon could find a happy medium, between weakening the AA filter a bit so it blurs less, but not eliminate it entirely.