Rishi Sanyal: Fun little thought experiment:
If we go by sensorgen, the 1D-X's pixels have a full-well capacity (FWC) of 90,000. Since the pixels on this sensor are 7.5x larger, we can extrapolate that given similar sensor capabilities, the pixels on this sensor can hold ~675,000 photoelectrons.
Now, since each doubling of ISO halves the FWC, ISO 4,000,000 will yield a FWC of roughly 675,000/40,000 = 16.875. Let's be generous and round that to 20. That means white is made from 20 photons.
If we generously place middle grey at 3 EV below clipping, that'd mean midtones are made from 20/8 = 2.5 photons, which itself yields a signal with SNR of 2.5/sqrt(2.5) = 1.6, which is already below most reasonable DR cutoffs. In other words, you'll have ~3 EV dynamic range at best, assuming no read noise whatsoever (bad assumption).
So, either my calculations are *way* off, or there's a limit to these insane ISOs. :)
Rishi, it would be fairly simple to write a script in Matlab to read in an image and simulate a Poisson process. What would be really cool (and something I've been wanting to do for a while but haven't had time) would be to shoot a whole load of ISO 100 images of the same scene and stack them to create a "noise free" master to use to simulate a perfect photon counting sensor.
You could then package that data back into a RAW file (difficult but by no means impossible) and distribute it along with actual RAW files shot of the same scene at different ISO settings. You'd then get a pretty good sense of how close we are to the best possible performance, and what that would mean for the malleability of the RAW data.
I suspect the "ideal" data would look an awful lot better despite only have a stop or two less noise amplitude, as the noise grain would be finer with no clumping or other artifacts.
20 photons per pixel potentially usable, provided you have ABSOLUTELY ZERO electrical noise. The middle image of this collage shows a simulation of 10 photons per pixel.
Chances of this sensor having zero noise.... zero. Add 3e of read noise into the mix and it's not going to look nice at all.
Kevin DiOssi: With the release or announcement of any Canon product these days, it's filled with experts from the interwebs.
Rather than taking a step back and imaging the practical uses for such a camera, and acknowledging that this product is one of the most innovative creations we've ever seen in cameras, we get the "but it doesn't have a Sony sensor" and "it would have more dynamic range if it didn't have a Canon sensor in it" quotes.
It's sad how poisonous DxO has become to the enthusiast and entry level amateur photographers. They really are ignorant to what's important thanks to a meaningless scoring system and scenarios of photography that are rarely ever exploited.
Just what exactly makes this "one of the most innovative creations we've ever seen in cameras"?
There really is no huge challenge in creating a super super high ISO setting on a camera, it's just gain applied to the analogue signal from the sensor.
You could have a ten billion ISO sensor - doesn't mean it would be any good. ISO is a measure of how quickly the highlights clip. It is not a measure of noise or anything else.
I'm not attacking Canon, I'd say the same if any manufacturer announced the same product. You simply can't get past shot noise, as Rishi mentioned above.
janist74: Looks great and high ISO improvements are welcome, but why no native ISO 100? This (a really clean base ISO) would be essential for landscape photographer and all APS-C can do it without a problem.
ISO is a measure of *sensitivity*, not *noise*.
Take a camera with native sensitivity of ISO 200. Glue a 1-stop ND filter to the sensor. Now it has a native sensitivity of ISO 100, but you haven't done anything to improve noise performance - it's still the same sensor.
nitroman: I downloaded both the raw files from Canon 5Ds and Canon 5Dsr.
It seems that if i apply more sharpening to the Canon 5Ds raw file, I get much the same results as the 5Dsr but without aliasing ... What does everyone else find ?
Also, if Canon add the anti aliasing filter using in camera software / firmware (rather than hardware modification), why don't they give us all the option of swicthing the anti aliasing on or off as required. This would then mean we only need one camera model not two ... ;)
Maybe the clever Magic Lantern guys can crack the code to do this ... lol
Deconvolution can restore the effects of an anti-aliasing filter (to an extent), but that's not what I was talking about.
I was talking about reversing the effects of aliasing (moire) itself, which is not possible, otherwise manufacturers would do it in software rather than hardware and there would be no D800e or 5DsR option.
For example there's no way to know whether a stripy pattern in the image is the result of moire, or whether the subject itself was stripy!
Canon don't implement the anti-aliasing filter using software. It's impossible to do this in the general case - once aliasing has occurred there's no guaranteed way to reconstruct the original signal.
SimenO1: Is it april 1. or is DxO killing all objectiveness in sensor testing?
I'm also chocked that they choose to manipulate the score on their own product by image stacking.
DxO, you wont work as a neutral source to sensor tests anymore to me.
Even with transparency there is still a clear conflict of interest in reviewing one's own products!
Have you seen the bag they've released with it?
It has a hole for the lens to poke through, turning it into the world's least convincing spy camera ;)
steelhead3: I see a court case coming up with Ricoh
The Canon D60 and Nikon D60 managed to peacefully co-exist without any lawsuits, I don't think anyone is seriously going to confuse the Leica Q and Pentax Q.
BeaverTerror: Mirrorless is the official term now? So instead of choosing a term which describes the camera, they have moronically chosen a term describing what the camera does not have. One day DSLRS will be a nostalgic memory and we will still be stuck with "mirrorless". Same situation as clipless bicycle pedals.
"If you have more imagination and skill, please suggest a better name."
Compact System Camera.
Unlike "mirrorless" is describes exactly what the camera is, after all my phone doesn't have a mirror, neither do most point and shoots...
Zeisschen: It always takes some time for the mass of people to understand DSLR isn't the latest & greatest anymore. It will always be there, but it will be a niche product for some people who really need an OVF, that's aout it.Times are changing now, they do since 3-4 years. I'm very happy to see companies like Sony and others (Olympus, Fuji, Samsung) innovating and simply change the status quo in the market, great times for us photographers!
AF aside, there's really no difference in IQ between say the Nikon D800 and Sony A7R, after all they use the same sensor. There may be minor differences in implementation but none of that can really be attributed to the form factor of mirrorless itself (unlike AF differences).
If Sony decide to cut Nikon out of the loop with the next sensor you might have to turn that argument around and say that mirrorless provides the highest IQ (unless you go to medium format).
arndsan: cool - do we get finally Leica reviews?
error = difference between compressed and uncompressed version
"You're tweaking of those Canon raws proved my point well enough."
The fact you picked out the small error and completely missed the much larger error seems to indicate that it was blind luck to me.
> The Sony A7 has Sony specific washed out colour trouble.
Only in your head...
LKJ: Android 4.4. The one that was last updated almost a year ago.
Good work Panasonic.
The first Android tablets came with an OS designed for phones that wasn't great for tablets , Google then rushed out a tablet specific version (3.0) for the Motorola Xoom whilst they worked on a version that could work well on phones and tablets (4.0), that's why 3.0 only existed for a short period. There was nothing wrong with it, it was just a stopgap.
DotCom Editor: The 5DS R resolution is 50.6 MP, much more than the Nikon D810 at 36.3 MP.
So, why does the ISO 100 RAW image from the Canon weigh in at 63.6 MB, while the D810 is 74.4 MB, according to the DPReview studio scene bulls-eye images?
Canon uses lossless JPEG compression in it's RAW files, Nikon has a totally uncompressed option which will naturally produce larger files even with a smaller pixel count.
HowaboutRAW: Good for Sony, now reconsider compressed raws in general.
Interestingly A is the version that was converted to 11-bit and D was the original lossless version. However in B was produced by taking the difference between A and D and exaggerating it to make it 4 times larger. If you can see the difference between A and D, the difference between B and D should have been glaring, so I'd say the results are pretty inconclusive.
My opinion is still that the differences are far too small (we're talking between 0 and 2 grey levels here for the most part) to produce any colour saturation effects, certainly if you examine the Hue Saturation Lightness values in photoshop they are identical between the lossy and lossless versions, hence I can't see how one could have "washed out colours" when the saturation is the same.
C was just D converted to JPEG, so if there was any difference there it would be JPEG artifacts.
You're clearly seeing what you expect and what you want to see not what's actually there, a classic case of confirmation bias. I shouldn't have labelled the images.
Here's a better test, no JPEG compression this time, look at the following images on your 10-bit monitor and tell me if any of them suffer from washed out colours:
You don't see any differences because there are no differences there (there are a few pixels that are 1 or 2 levels different, but that's all).
As promised I've added a routine to the dcraw RAW converter to apply Sony's lossy compression curve to convert the data from 14 bits to 11 bits and back prior to any other processing. Here is a standard conversion of a 14-bit lossless Canon RAW:
And here it is the same file after lossy compression / decompression:
Please let me know if you can see any difference or washed out colours.
If you're interested, here is the source code:
And here's a windows executable for you to try yourself (use the -L switch to activate lossy compression):