The Mirrorless Revolution

[Hans Kruse]

I don't think that is necesarily true Hans.
DXO mark shows a graph of the percentage of the maximum sensor resolution tested (the so-called percentage of perceptual MPix)

This is what they write about it on their site:
Perceptual MPix: a much simpler tool to score and compare lenses

P-Mpix is the unit of a sharpness measurement. The number of P-Mpix of a camera/lens combination is equal to the pixel count of a sensor that would give the same sharpness if tested with a perfect theoretical optics, as the camera/lens combination under test.

For example, if a camera with a sensor of 24Mpix when used with a given lens has a P-Mpix of 18MPix, it means that somewhere in the optical system 6Mpix are lost, in the sense that as an observer you will not perceive the additional sharpness that these 6Mpix should have added to the photos if everything was perfect.

In other words it indicates the ability of the lens and other optical components of a camera to utilize, from a visual perspective, the number of pixels of the camera sensor.  P-MPix expresses the result using a figure that can easily be compared to the camera sensor’s MPix figure to show the quality of the lens.

Since the A7R is a 36 MP camera and the 5DIII a 22 MP camera this means that a 60% score on the A7R/lens combo 0.6 x 36 = 21,6 million "effective" pixels
A 70 % score on the 5DIII means the lens/sensor combination is using 0.7 x 22 = 15.4 million "effective" pixels
So 60% of the A7R score still has a higher resolution in the final resulting raw file then a 70% score of the 5DIII

This doesn't mean the Canon 24-70/2.8 is a less sharp lens then the Sony/CZ 24-70/4, it's only what DXO measures and reports for the lens/sensor combination of both cases.

To draw any conclusions on the lenses alone you would need to mount the Canon lens on the Sony body (with an adapter) and do a side-by-side comparison on the same sensor.

here's a link to the full story on the DXOMark website

I find it really hard to find detailed descriptions of what DxO curves mean in detail. But as far as I can interprete the curves of acutance this is not a percentage of MP (and not M-Pix)

See here the same lens on two difference cameras, D800E and 5D mkIII. The Sigma 35 f/1.4 lens. This shows to me clearly that this lens is able to resolve more on the D800E than on the 5D mkIII. Therefore, if two different lenses on the D800E and the 5D mkIII has curves overlaying each other then the lens on the Canon is superior to the lens on the Nikon.

[pegelli]

Hans, you might be right and browsing the DXO site more I came to the conclusion that the % acutance in the graphs are most likely not a % of MPix.

However DXO doesn't provide any information what the measurement really is, and even in their forum are very vague about providing more details see here. They promised an article on it for early 2014, but in June 2014 (and up to now) it remains unpublished. I hope they publish it soon, because I think a lot of people are interested in it (I am at least one and I guess you too )

Their description says it's some kind of perceptual measure of visual details in standard test chart, but not sure yet how it varies from sensor to sensor and if there is an impact of sensor megapixels or not.

[Guillermo Luijk]

It's sometimes better to dig into the details where you see that the DR of the Canon is a problem at low ISO's but from about ISO 400 it changes sides. The noise characteristics is way better on FF as you can see.

Canon's DR at low ISO is not generally a problem, in fact it's very good (it's the best Canon FF sensor ever). M4/3 is simply a bit better having 1/4 of sensor surface. That could be seen as a technological humilliation but that's another story.

The noise characteristics you mention must be first understood: SNR at 18% RAW exposure is always good on any sensor. 5D's is simply a bit better than the M4/3 but that difference is irrelevant since both are more than good enough (SNR>36dB for the M4/3 is excellent, invisible noise!). Noise is much better measured by the DR plot, which represents noise when lifting the shadows (where noise can really be a problem, and not at 18% of sensor saturation):




And here the M4/3 beats the 5D at low ISO. That makes the M4/3 a more balanced sensor at low ISO (noise is not so good in well exposed areas but better in the shadows). As a photo enthusiast and hobbist, not a professional, my choice is clear, since I own this beauty:




I don't even think of carrying the bulky 5D+glass with me, just common sense. I only miss sometimes the ability to get shallower DOF, which is mostly solved using luminous primes that work fine right from max. aperture on M4/3.

Regards

[Hans Kruse]

Canon's DR at low ISO is not generally a problem, in fact it's very good (it's the best Canon FF sensor ever). M4/3 is simply a bit better having 1/4 of sensor surface. That could be seen as a technological humilliation but that's another story.

The noise characteristics you mention must be first understood: SNR at 18% RAW exposure is always good on any sensor. 5D's is simply a bit better than the M4/3 but that difference is irrelevant since both are more than good enough (SNR>36dB for the M4/3 is excellent, invisible noise!). Noise is much better measured by the DR plot, which represents noise when lifting the shadows (where noise can really be a problem, and not at 18% of sensor saturation):




And here the M4/3 beats the 5D at low ISO. That makes the M4/3 a more balanced sensor at low ISO (noise is not so good in well exposed areas but better in the shadows). As a photo enthusiast and hobbist my choice is clear: since I own this beauty:




I don't even think of carrying the bulky 5D+glass with me.

Regards

How can the m4/3 be better in the shadows when the SNR is not on the level of the 5D III? I'm not referring to deep shadows where the 5D III starts with the banding. The noise level must be amplified equally when pushing shadows in post processing and therefore the sensor with the lowest SNR should win. Where is the flaw in that argument?

When shooting landscapes at ISO 100 and there are very bright highlights I bracket the shots and choose the most exposed shot in Lightroom where only non essential highlights are clipped. That means that I need to push shadows and this is done via the shadows slider as well as grad filters. What I find with the 5D III is that there is a level of the shadows where it starts banding, but before that level the shadows are fine with low noise. Clearly the Nikon D800E which I also shoot has a higher tolerance in the shadows due to the larger DR, but as long as the shadows are not breaking the 5D III I find little difference in the shadows.

Since the m4/3 is a smaller sensor if the pictures are resampled to the same size, the m4/s should show more noise in the shadows after this level of post processing that I just described. That's my thinking and if the logic is wrong, then where does it fall apart? Besides that I wish the Canon sensors will be better in the future.

[Hans Kruse]

Hans, you might be right and browsing the DXO site more I came to the conclusion that the % acutance in the graphs are most likely not a % of MPix.

However DXO doesn't provide any information what the measurement really is, and even in their forum are very vague about providing more details see here. They promised an article on it for early 2014, but in June 2014 (and up to now) it remains unpublished. I hope they publish it soon, because I think a lot of people are interested in it (I am at least one and I guess you too )

Their description says it's some kind of perceptual measure of visual details in standard test chart, but not sure yet how it varies from sensor to sensor and if there is an impact of sensor megapixels or not.

So you agree on my conclusions on the Canon lens in my previous post?

And thanks for the link to the forum as I have been searching also for clear descriptions and not vague ones. I though that my lack of clear understanding perhaps was since I'm not on the expert level of the DxO folks who do these measurements and calculations. So therefore I took the reverse engineering route to find consistency in the graphs. I find the information very valuable when evaluating a new lens purchase although there will be sample variations that the graphs don't show.

[ErikKaffehr]

Hi,

I have the SAL 24-70/2.8 ZA, what I can say is that it is very sharp in the center but needs to be stopped down to f/8 - f/11 for accreptable corners.

I have seen some real MTF tests, made at Hasselblad for both the Canon 24-70/2.8 and the ZA. In that case the Canon was better. I am actually considering to replace the 24-70/2.8 with a Sigma 24-105/4, but that lens seems to have some issues, too. But, that must also be said, I am quite happy with the 24-70/2.8 ZA, many great images coming from that lens.

Best regards
Erik

So you agree on my conclusions on the Canon lens in my previous post?

And thanks for the link to the forum as I have been searching also for clear descriptions and not vague ones. I though that my lack of clear understanding perhaps was since I'm not on the expert level of the DxO folks who do these measurements and calculations. So therefore I took the reverse engineering route to find consistency in the graphs. I find the information very valuable when evaluating a new lens purchase although there will be sample variations that the graphs don't show.

[Guillermo Luijk]

How can the m4/3 be better in the shadows when the SNR is not on the level of the 5D III? I'm not referring to deep shadows where the 5D III starts with the banding. The noise level must be amplified equally when pushing shadows in post processing and therefore the sensor with the lowest SNR should win. Where is the flaw in that argument?

When shooting landscapes at ISO 100 and there are very bright highlights I bracket the shots and choose the most exposed shot in Lightroom where only non essential highlights are clipped. That means that I need to push shadows and this is done via the shadows slider as well as grad filters. What I find with the 5D III is that there is a level of the shadows where it starts banding, but before that level the shadows are fine with low noise. Clearly the Nikon D800E which I also shoot has a higher tolerance in the shadows due to the larger DR, but as long as the shadows are not breaking the 5D III I find little difference in the shadows.

Since the m4/3 is a smaller sensor if the pictures are resampled to the same size, the m4/s should show more noise in the shadows after this level of post processing that I just described. That's my thinking and if the logic is wrong, then where does it fall apart? Besides that I wish the Canon sensors will be better in the future.

SNR varies differently with RAW exposure between cameras, they only share the following behaviour: the higher the RAW levels the higher the SNR.

Near saturation (what DxO SNR18% benchmark measures), the 5D3 has better SNR than M4/3 sensors. You must understand 18% is roughly 2,5 stops from saturation so there is little interest in knowing how noisy a sensor is there (they're all good enough).

On the opposite, in the shadows (what DxO DR benchmark measures), the M4/3 has better SNR than the 5D3. So M4/3 has less shadows noise than the 5D3, the real challenge of sensors regarding noise.

Regards

[Hans Kruse]

SNR varies differently with RAW exposure between cameras, they only share the following behaviour: the higher the RAW levels the higher the SNR.

Near saturation (what DxO SNR18% benchmark measures), the 5D3 has better SNR than M4/3 sensors. You must understand 18% is roughly 2,5 stops from saturation so there is little interest in knowing how noisy a sensor is there (they're all good enough).

On the opposite, in the shadows (what DxO DR benchmark measures), the M4/3 has better SNR than the 5D3. So M4/3 has less shadows noise than the 5D3, the real challenge of sensors regarding noise.

Regards

Ok, so the SNR measurements by DxO is pretty much useless from what you say. So what we really should be looking at SNR as a function of exposure. But the DR measurements is not that. When I lift exposure in post processing when shooting at ISO 100 i digitally amplify the signal (including base noise aka. read noise from the AD converter) and not analogue as in the camera from which the DR measurements are derived. For an ISO less sensor I understand that there is little to no difference in the two but for the Canon sensor which is not ISO less in nature the difference is quite significant. So are there any source from where the SNR curve as a function of exposure could be derived to we properly could compare the Canon and the Oly camera? Am I right here in these considerations?

[Hans Kruse]

Hi,

I have the SAL 24-70/2.8 ZA, what I can say is that it is very sharp in the center but needs to be stopped down to f/8 - f/11 for accreptable corners.

I have seen some real MTF tests, made at Hasselblad for both the Canon 24-70/2.8 and the ZA. In that case the Canon was better. I am actually considering to replace the 24-70/2.8 with a Sigma 24-105/4, but that lens seems to have some issues, too. But, that must also be said, I am quite happy with the 24-70/2.8 ZA, many great images coming from that lens.

Best regards
Erik

Regarding the Sigma 24-105 f/4 OS you know I have this lens for the Nikon D800E and have not tested it for landscapes yet for the D810 (saying this because Lloyd Chambers found some odd softness in some cases with the D800E compared to the Canon 5D III in his tests). I found it mostly very good, but with some smaller areas with some softness. So not ideal but a very nice lens still and a very good range for a walk around lens. For those who are willing to carry it and the camera Combined it is a bit heavy.

[MoreOrLess]

The big issue with "normal" signal noise at base ISO for me would be contrast processing(especially for monochrome conversions), I found my old 550D lacking in that respect and even shooting FF can show issues if you push hard enough.

[Guillermo Luijk]

Ok, so the SNR measurements by DxO is pretty much useless from what you say. So what we really should be looking at SNR as a function of exposure. But the DR measurements is not that. When I lift exposure in post processing when shooting at ISO 100 i digitally amplify the signal (including base noise aka. read noise from the AD converter) and not analogue as in the camera from which the DR measurements are derived. For an ISO less sensor I understand that there is little to no difference in the two but for the Canon sensor which is not ISO less in nature the difference is quite significant. So are there any source from where the SNR curve as a function of exposure could be derived to we properly could compare the Canon and the Oly camera? Am I right here in these considerations?

SNR is a function of exposure and ISO. In fact DxO used to provide the SNR curves, but unfortunately they don't do it anymore:

• If you look at that graph, the SNR 18% plot is the slice for RAW level (x-axis)=18%. This info. could be useful for very purist studio shooters to find out how hyper clean their images can be, but IMO is quite useless.
• The DR plot on the contrary represents the number of stops between saturation and those deep shadows where SNR falls below 0dB. This gives a good idea of the RAW exposure range acceptably free of noise to distinguish textures. However SNR=0dB is a completely ruined by noise image, so it's only valid as an engineering DR criteria and to stablish comparisions between cameras. As a practical rule, DxO DR figures can be converted to realistic DR figures just by substracting 2 stops to their provided DR (this is equivalent to a SNR=12dB criteria, a good choice IMO for the maximum acceptable noise).

Regarding the different behaviour of iso-less sensors vs Canon sensors, this can be well understood again by looking at the SNR plots:



K5's iso-less sensor doesn't improve SNR by pushing ISO while aperture/shutter remain constant, that is why the thick black line remains at constant SNR. On the contrary de 5D2 sensor benefits from pushing ISO increasing its SNR, that is why its thick black line gets higher SNR the higher the ISO. This makes pushing ISO on Canon sensors much more necessary/useful than in Sony sensors when aperture/shutter don't suffice to obtain high enough RAW levels.


Regarding Canon sensor's DR, they seem to have prioritized other areas in the last years (video?):


DR of Canon/Nikon APS sensor vs time

Regards

[pegelli]

So you agree on my conclusions on the Canon lens in my previous post?

Let's say that for me the jury is still out, you might be right or not depending on the definition of the % acutance DxO will hopefully give us one day. So I'm really hoping for a good explanation of what that number means.

The best test would obviously be to test the Canon lens on the same a7R body (pretty easy to do with the right adapter), when DxO would provide that the discussion can be closed without knowing how the percentage is actually defined.

[Hans Kruse]

Let's say that for me the jury is still out, you might be right or not depending on the definition of the % acutance DxO will hopefully give us one day. So I'm really hoping for a good explanation of what that number means.

The best test would obviously be to test the Canon lens on the same a7R body (pretty easy to do with the right adapter), when DxO would provide that the discussion can be closed without knowing how the percentage is actually defined.

Here is another example of the same lens tested on different bodies and all with no or compensated OLPF (D800E): Nikon D610, D800E and D7100. There are not yet any measurements for the D810.

So to me it is clear that the curves are absolute such that the higher % means a higher resolution. But I would like you like some clearer definitions and ideally also how various things are calculated. What are the formulas? I do realize that I probably would not understand all the formulas anyway

[barryfitzgerald]

Not sure why we've wandered onto dynamic range, as I was using a 5dMkIII a bit I can comment though, very nice camera but DR wise yes it clearly lags rivals, and it's not as good as some crop sensor cameras in that dept, DR has only marginally increased with Canon over the years.

However the main problems that face mirrorless remain, price relative to other normal crop sensor and FF bodies, and the system.
People can point to the new wave of modern designs, but these come with a higher challenge for optics (with the lens sitting closer to the sensor) this was and remains a problem for designers with distortion and fall off much more of a problem than normal DSLR type designs.

ILC's are interesting, but they are not going to represent the future for many buyers simply because they don't offer a compelling reason to move to them. And looking at some of the prices makers are charging, the camera industry is frankly fooling itself. In the last decade there has been a huge reduction in the price of consumer electronics, from computing to televisions, audio equipment and other items. Yet the camera industry feels it can remain immune to this, and they are horribly wrong... this is one of the reasons for the decline in sales of cameras with makers desperately trying to get premium prices (on higher end compacts and cameras) when the market simply isn't going to take it. We've loads of cameras, and most people won't be updating very often, and until the industry works out that they're overpriced, they won't get the sales on other items to make the profits.

[MoreOrLess]

Not sure why we've wandered onto dynamic range, as I was using a 5dMkIII a bit I can comment though, very nice camera but DR wise yes it clearly lags rivals, and it's not as good as some crop sensor cameras in that dept, DR has only marginally increased with Canon over the years.

However the main problems that face mirrorless remain, price relative to other normal crop sensor and FF bodies, and the system.
People can point to the new wave of modern designs, but these come with a higher challenge for optics (with the lens sitting closer to the sensor) this was and remains a problem for designers with distortion and fall off much more of a problem than normal DSLR type designs.

ILC's are interesting, but they are not going to represent the future for many buyers simply because they don't offer a compelling reason to move to them. And looking at some of the prices makers are charging, the camera industry is frankly fooling itself. In the last decade there has been a huge reduction in the price of consumer electronics, from computing to televisions, audio equipment and other items. Yet the camera industry feels it can remain immune to this, and they are horribly wrong... this is one of the reasons for the decline in sales of cameras with makers desperately trying to get premium prices (on higher end compacts and cameras) when the market simply isn't going to take it. We've loads of cameras, and most people won't be updating very often, and until the industry works out that they're overpriced, they won't get the sales on other items to make the profits.

I'm not sure that the sectors of the camera industry we discuss is really all that similar to most of what we'd call "consumer electronics", even an entry level ASPC DSLR's is actually quite an exotic piece of kit that's only going to be desired by a relatively small part of society compared to a flat screen TV or a smartphone. Your basic compact camera is more on the level of typical consumer electronics and that had indeed come down greatly in price to the extent that I suspect that even without the competition from smartphones profit margins are very low.

Your point about optics does for me highlight a bit of a misrepresentation of SLR tech in the digital age as some kind of leftover that's been clung too purely because it was established. In reality I would say that SLR tech persisted because it was actually a very good match for digital sensors. As you say retrofocal lens designs generally suit digital sensors, mirrors don't need any power to run on already power hunger devices, batteries used to provide that power can fit into grips needed to handle the lens sizes needed to deliver ever increasing performance demands. Even when it comes to AF your dealing with fundamentally different setups with no guarantee that tracking performance being matched by mirrorless will meet existing DSLR standards quickly or easily.

I think the just announced D750 is quite significant as well. I'd seen talk in the past that things like widely spaced professional AF and 100% viewfinders could not be fitted into smaller FF DSLR bodies but this camera does both. It makes the D800 an unrealistic comparison for the long term situation since I see no reason a body the size of the D750 couldn't include a higher resolution sensor.

[ErikKaffehr]

Hi,

SLRs have been around for 80+ years, mirrorless is quite new.

The way I see it, SLRs were invented to achieve "what you see is what you get" representation of the image. The cost was complexity, adding a swinging mirror that is of course causing a few vibration problems and also induces an alignment problem.

With the arrival of AF we added a secondary mirror and even more alignment problems.

With modern sensors, we can use the actual signal of the sensor to build a preview image. We get "what you see is what you get" without the complexity of two swinging mirrors.

Removing the mirror makes it possible to reduce the flange distance, thus making the camera smaller. The reduced flange distance is really an option, you can still use DSLR lenses on mirrorless.

But, mirrorless has a long way to go. They will be much better after a few years of development.

Best regards
Erik

Not sure why we've wandered onto dynamic range, as I was using a 5dMkIII a bit I can comment though, very nice camera but DR wise yes it clearly lags rivals, and it's not as good as some crop sensor cameras in that dept, DR has only marginally increased with Canon over the years.

However the main problems that face mirrorless remain, price relative to other normal crop sensor and FF bodies, and the system.
People can point to the new wave of modern designs, but these come with a higher challenge for optics (with the lens sitting closer to the sensor) this was and remains a problem for designers with distortion and fall off much more of a problem than normal DSLR type designs.

ILC's are interesting, but they are not going to represent the future for many buyers simply because they don't offer a compelling reason to move to them. And looking at some of the prices makers are charging, the camera industry is frankly fooling itself. In the last decade there has been a huge reduction in the price of consumer electronics, from computing to televisions, audio equipment and other items. Yet the camera industry feels it can remain immune to this, and they are horribly wrong... this is one of the reasons for the decline in sales of cameras with makers desperately trying to get premium prices (on higher end compacts and cameras) when the market simply isn't going to take it. We've loads of cameras, and most people won't be updating very often, and until the industry works out that they're overpriced, they won't get the sales on other items to make the profits.

[deejjjaaaa]

But, mirrorless has a long way to go. They will be much better after a few years of development.

there is only __one__ item where dSLM still behind dSLR = tracking in CAF

[deejjjaaaa]

I'd seen talk in the past that things like widely spaced professional AF and 100% viewfinders

dSLM do focus @ EV-4 in ___any point in frame__ - that is a what ___widely spaced professional___ AF is.

[ErikKaffehr]

Hi,

I have no issue with that view, but it may be possible on sensor phase detection already takes us there.

I see some other issues mirrorless:

1) Lag, mirrorless needs to read out the viewfinder image from the sensor which introduces a lag. SLRs also have a lag, due to mirror flip up, but the image in the viewfinder is real time. I don't know how important this is. I am not sure mirrorless is good enough for sports.

2) Viewfinder image contrast, but this amy be tuneable.

3) Viewfinder resolution

I would expect significant advances in all those areas.

Best regards
Erik

there is only __one__ item where dSLM still behind dSLR = tracking in CAF

[deejjjaaaa]

ILC's are interesting, but they are not going to represent the future for many buyers simply because they don't offer a compelling reason to move to them.

most buyers (folks outside LuLa) do buy just a camera and kit lenses... so if & when C&N decide to finally move in there will be a compelling reason - because all "rebels" will be dSLMs.