The best FF sensor

[ProfHankD]

By now, I'm not saying anything people haven't heard: the Sony A7RII has the best FF sensor you can currently buy.

Actually, it has a bunch of minor improvements over the A7II... but not much compelling to me. That's why I took so long to get one. I could easily mistake one for the other both in handling and in image quality (as long as I'm not viewing the image too large for 24MP). Aside from a 42MP sensor that's great in low light, the video support is probably the biggest step up -- and it can do excellent 4K or 1080 at 120FPS (which is a big deal for some of the stuff I'm doing in my research). I should also mention that I shoot manual lenses nearly 100% of the time and it's great for that, but AF is still quite effective on the A7RII -- including AF of manual lenses via a TechArt LM-EA7.

So, do I recommend the A7RII? Heck yeah! However, Sony has a bunch of body options I can recommend just as strongly given their price points. Pick the cheapest that does what you need... and if you need everything the best, this is the one.

BTW, for social flash photography you're better off with a smaller camera that has a self-contained flash; my low score on that is just because of the bulk of camera and flash.

PS: I own and use more Canon cameras than Sonys, but just because of CHDK and ML; the Sonys are better in nearly every way, just not as programmable. Now that Open Memories is starting to make real progress in allowing Sonys to be programmed....

 

[ProfHankD]

Of course, once the sensor pixel density gets enough past lens resolution, everything works without an AA filter because the lens itself acts as a low pass filter.

Not to mention the 100% effective fill factor, which is now pretty common.

I didn't mention it. ;-) However, there's still the Bayer CFA pattern causing the sampling to fail the usual Nyquist criteria... but we will not mention that here either. ;-)

As I use my lenses most of the time, I think the A7RII would be ok with/without an AA filter. ...

I think we're in material agreement here, Hank. An AA filter, particularly a weak one like the ones on the rest of the a7x line, wouldn't hurt things much in my estimation, though, and would IMHO net help.

Absolutely, Jim. I think this was probably more of a marketing-driven decision for Sony -- your highest-resolution camera shouldn't have an AA filter because that makes it appear to have slightly lower sharpness.

OBTW, why are the a7x AA filters anisotropic? Anybody know?

Darn good question. My guess would be that perhaps it has to do with the Bayer CFA and trying to reduce the effect I certainly didn't mention several lines above? ;-)

 

[OlyPent]

16 bit medical grade ccd. Consumer cmos = too crude

 

[JimKasson]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

 

[Abrak]

In low light, I would prefer the Exmor HD sensor to have bigger pixels, such as in the a7 or a7s.

I have never been that convinced by the 'larger pixel' argument for low light. I find that the 'extra resolution' of the 'small pixel' A7rii tends to mean I can trade some 'detail' for 'noise reduction' with the A7rii relative to the A7ii (in addition to the A7rii's sensor's natural advantages.)

 

[OlyPent]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

 

[vyoufinder]

Note that I said "for normal people" ... $40,000 for a camera isn't normal people .

 

[crosstype]

IMX251 is the best FF sensor for short exposure, but a poor choice for long exposure.

IMX094 is the best FF sensor for long exposure and dynamic range.

 

[OlyPent]

Note that I said "for normal people" ... $40,000 for a camera isn't normal people .

I knew one guy who had an 80mp medium format and I know people who have astro cameras in that price range. But, yes; it's not common.

 

[jonas ar]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

--
http://blog.kasson.com

Uneven response, dead and or hot pixels.

I'm using the Andor neo 5.5 sCMOS and Hamamatsu ORCA flash 4.0 CMOS cams for light sheet microscopy, primarily imaging fluorescent probes in the far red and NIR range. They are sufficiently refined for that purpose

 

[OlyPent]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

--
http://blog.kasson.com

Uneven response, dead and or hot pixels.

I'm using the Andor neo 5.5 sCMOS and Hamamatsu ORCA flash 4.0 CMOS cams for light sheet microscopy, primarily imaging fluorescent probes in the far red and NIR range. They are sufficiently refined for that purpose

Just curious, what kind of things are probed in that application?

 

[jonas ar]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

--
http://blog.kasson.com

Uneven response, dead and or hot pixels.

I'm using the Andor neo 5.5 sCMOS and Hamamatsu ORCA flash 4.0 CMOS cams for light sheet microscopy, primarily imaging fluorescent probes in the far red and NIR range. They are sufficiently refined for that purpose

Just curious, what kind of things are probed in that application?

We are for example doing whole mouse brain imaging probed with various antibodies (iDISCO protocol) or examining the brain distribution of fluorescently tagged peptides following peripheral administration.

 

[ProfHankD]

IMX251 is the best FF sensor for short exposure, but a poor choice for long exposure.

This is the sensor in the A7RII, RX1RII, and A99II.

IMX094 is the best FF sensor for long exposure and dynamic range.

The sensor in the A7R, D800, and D810.

Are you saying the IMX094 has greater well capacity and/or lower leakage?

I would accept that the IMX094 is better in that PDAF pixels screw-up a variety of things (mostly just near where they are), but I haven't seen anything that implies it is better in terms of performance per unit sensor area. Just to be clear, I don't consider it a fair comparison to compare per pixel on sensors with different pixel densities.

 

[DFPanno]

Yes, great sensor.

The fact that it is not covered with an AA filter certainly contributes to the increased acuity one sees in its images.

Personally I find the A7R IIs images to be distinctly better than the A7 II's I had for a bit. I wonder to what degree the filter contributes to that difference.

My experiences with the D800/D800E convinced me that an AA filter is a good thing for most purposes, and that a little deconvolution can compensate for the MTF falloff in the 0.2 to 0.4 cy/px region.

I've also done a lot of simulations of cameras with and without various kinds of AA filters, that reinforce my impressions.

Of course, the whole topic is subject-dependent. A clothing photographer may love AA filters, while a rural landscape photographer my hate them.

AA filters are clearly advantageous with some subjects. ...

You are certainly right about the ability of knowledgable post-processing to virtually negate any disadvantage intrinsic to a filter.

Whenever I think of resolution, I think in terms of Nyquist sampling. I'll skip my long-winded explanation here and simply say that most AA filters should really be much stronger than they are, but humans tend to like the extra detail even if it's potentially wrong detail, as long as it doesn't look artificial. Of course, once the sensor pixel density gets enough past lens resolution, everything works without an AA filter because the lens itself acts as a low pass filter.

In my experience, few legacy lenses resolve high enough wide open to desperately need an AA filter at the pixel density of a 24MP APS-C sensor. However, stopped down to f/8, many lenses do. In that context, I believe the A7RII's 42MP FF sensor (18MP APS-C density) could really use an AA filter for the center of the frame. Most lenses drop in resolution off axis; in the corners, very few of my lenses shooting wide open would need an AA filter.

As I use my lenses most of the time, I think the A7RII would be ok with/without an AA filter. Serendipity also plays a role here relative to the scenarios Jim noted. When I'm shooting people with moire-prone clothing, I'm often shooting with the lens wide open; I tend to shoot landscapes stopped down. Thus, I'm actually using the lens as a bit of a variable AA filter.... ;-)

Interesting.

I thought that as the "frequency" of sensor sites was raised it would exceed the "frequency" of most things found in the world and as such would generally not induce moire.

Is the lack of the filter just marketing then?

 

[ProfHankD]

I thought that as the "frequency" of sensor sites was raised it would exceed the "frequency" of most things found in the world and as such would generally not induce moire.

It does, but the world's resolution is incredibly high. It is really limited by lens resolution, which can still exceed Nyquist for a 42MP FF sensor -- especially taking the Bayer CFA into account.

Is the lack of the filter just marketing then?

Can't say that, but marketing is certainly a major reason to drop the AA before the pixel density really justifies dropping it. Alternatively, one could simply be computing the cut off assuming a poorer-resolving lens... of which there are plenty.

 

[DFPanno]

I thought that as the "frequency" of sensor sites was raised it would exceed the "frequency" of most things found in the world and as such would generally not induce moire.

It does, but the world's resolution is incredibly high. It is really limited by lens resolution, which can still exceed Nyquist for a 42MP FF sensor -- especially taking the Bayer CFA into account.

Is the lack of the filter just marketing then?

Can't say that, but marketing is certainly a major reason to drop the AA before the pixel density really justifies dropping it. Alternatively, one could simply be computing the cut off assuming a poorer-resolving lens... of which there are plenty.

Interesting; thanks !

 

[JimKasson]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

Uneven response, dead and or hot pixels.

CCD: blooming, high RN. Dark current issues, too -- LENR was invented for CCDs, and then ported over to CMOS.

Uneven response in CMOS? What are you talking about? I'm seeing PRNU below 0.2% with the a7RII.

Do you really want to go back to CMOS sensors? I'm glad those bad old days are behind us. I do have one camera witha CMOS sensor: the H2D-39; it's a noisy sucker, but OK if you can control the light well. If you hurry, you can pick up a Phase One camera with a CCD sensor, but do it fast, because they, like most everybody else, are switching to CMOS.

70 degree below ambient sensor cooling? Try to actually use that in the field and see how much condensation you get.

Jim

--
http://blog.kasson.com

 

[OlyPent]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

 

[ProfHankD]

Do you really want to go back to CMOS sensors? I'm glad those bad old days are behind us. I do have one camera witha CMOS sensor: the H2D-39; it's a noisy sucker, but OK if you can control the light well. If you hurry, you can pick up a Phase One camera with a CCD sensor, but do it fast, because they, like most everybody else, are switching to CMOS.

I agree with you, Jim -- CCDs are not the current best answer for "normal" cameras. Aside from any IQ issues, slow readout is a huge problem. However, CCDs are far from dead. For example, the lower-end Canon PowerShots all use CCDs, and there are some situations where CCDs actually do perform better.

Anyway, an interesting overview of CCD vs. CMOS is at this Teledyne page . It's a very fair-sounding treatment that sums up the relevant issue here as:

"CMOS area and line scan imagers outperform CCDs in most visible imaging applications."

 

[JimKasson]

Do you really want to go back to CMOS sensors? I'm glad those bad old days are behind us. I do have one camera witha CMOS sensor: the H2D-39; it's a noisy sucker, but OK if you can control the light well. If you hurry, you can pick up a Phase One camera with a CCD sensor, but do it fast, because they, like most everybody else, are switching to CMOS.

I agree with you, Jim -- CCDs are not the current best answer for "normal" cameras. Aside from any IQ issues, slow readout is a huge problem. However, CCDs are far from dead. For example, the lower-end Canon PowerShots all use CCDs, and there are some situations where CCDs actually do perform better.

Anyway, an interesting overview of CCD vs. CMOS is at this Teledyne page . It's a very fair-sounding treatment that sums up the relevant issue here as:

"CMOS area and line scan imagers outperform CCDs in most visible imaging applications."

Good article, Hank. One thing not mentioned is that in applications where the sensors need to be physically large and the pitch needs to also be large (like 10 or 15 um), like some telescope sensors, CCD has a cost advantage because the fabs and the per-unit-area fabbed chip cost is lower for large sensors.

It's interesting. I went to the Fairchild web site to get an example of a big sensor, and it looks like they have discontinued all their CCD standard products, although they will still do custom CCDs.

Jim

 

[JimKasson]

16 bit medical grade ccd. Consumer cmos = too crude

What's crude about CMOS?

Uneven response, dead and or hot pixels.

CCD: blooming, high RN. Dark current issues, too -- LENR was invented for CCDs, and then ported over to CMOS.

Uneven response in CMOS? What are you talking about? I'm seeing PRNU below 0.2% with the a7RII.

I think that, for photographic, as opposed to scientific purposes, when the PRNU drops to well below the photon noise, its precise value becomes irrelevant.

Jim