skibum5 wrote:
Keep in mind DR is mostly important when looking into deep shadows, above that the 18% SNR measurements of DxO are probably more sensible of a comparison and here the 5D2 does well over a stop better than all the APS-C from Canon at all ISO. I think like 1.6 stops better than the 40D, for instance.
Only when one normalizes the results to a common screen size.
At the pixel level, the 18% SNR results between 5D2 and 40D are pretty close.
Somehow, I was hoping the 7D2 could do that magical leap into the sensor technology stratosphere like the D90/D5000/D300s and D3x.
But it doesn't appear Canon is able to deliver on this front.
My 40D topped out at 11.32 but it was a true performance outlier, my 5D2 at 11.24 and my 50D at 11.07 and we have the 7D at 11.26 (actually I saw some data implying it might do better at ISO200 and this 11.26 is from ISO100).
A more fair comparison is normalized (here to 8.2 MP but it doesn't matter, ignore the absolute number and just look at relative differences):
5D2: 11.9 (ISO 100)
7D: 11.9 (ISO 100, ISO 200 might be a tiny bit better)
50D: 11.5 (ISO 200)
40D: 11.5 (ISO 200; my 40D was a real outlier, better than most)
for kicks at ISO1600:
5D2: 11.1
50D: 9.9
40D: 9.7
Keep in mind DR is mostly important when looking into deep shadows, above that the 18% SNR measurements of DxO are probably more sensible of a comparison and here the 5D2 does well over a stop better than all the APS-C from Canon at all ISO. I think like 1.6 stops better than the 40D, for instance.
alundeb wrote:
Nikon (in some cameras) use a very small Black Level offset.
By using a larger black level offset, Canon allows for recording all possible noise (and signal) even at very high ISO's.
I did the Nikon trick on the raw file, clipped at the Black Current level. Whooops.
ISO 100 Sigma 3.1, DR 12.1 stops
Normalized to 12MP: DR 12.4 stops
And Yes, even after the trick, we will detetct even such small amounts of light, but with a loss of detail.
That's quite a revelation.
If it's possible to detect even small amounts of light, why doesn't Canon resort to the same trick? At least it'll look good on DXOMark tests.
The loss of detail is minuscule IMO except for astro-photographers.
alundeb wrote:
Estimate of the photon full well (saturation) of the 7D
The background of the still life scene used by image-resource.com has some areas that are uniformly lit, and clean.
For high ISO's (6400-12800), the photon shot noise is practically all that contributes to the Standard Deviation (or variance, or Sigma) in those areas.
Sorry, it is at low ISO's that photon shot noise is all that contributes in bright areas. At high ISO there is substantial contribution from the std dev of pixel RAW values from the read noise, at any illumination level. In the top stop or two at base ISO there can be a potential contribution from pixel response nonuniformity, so either choose a midtone at base ISO or a bright area at ISO 200.
ejmartin wrote:
Sorry, it is at low ISO's that photon shot noise is all that contributes in bright areas. At high ISO there is substantial contribution from the std dev of pixel RAW values from the read noise, at any illumination level. In the top stop or two at base ISO there can be a potential contribution from pixel response nonuniformity, so either choose a midtone at base ISO or a bright area at ISO 200.
I get that in principle, but for this camera, the read noise at high ISO is extremely low ( see one of my later posts above).
In this case, the Std.dev at ISO 12800 is 288 and the read noise is 22.
The Std.dev at ISO 200 is 36 and the read noise is 4.7
Edit: Forgot to mention, that at low ISO, image detail will contribute a lot more to Std.Dev.
alundeb wrote:
Getting usable DR like the 40D is not bad, but I want a tiny bit of progress
Checked ISO 200, 1600 and 6400:
ISO 200: Sigma 4.7 , sat. 15300-2048, DR 11.46 DRn8 12.05
ISO 1600: Sigma 4.9 (!!!) DR 11.4 DR n8 11.99
ISO 6400: Sigma 11 DR 10.23 DR n8 10.82
Looks like they did a helluva job with the sensor Read Noise, and the amp/AD chain is still the bottleneck at low ISO.
Hmmm.........
I just got some of the IR images myself and i am not getting the same numbers so far for this (ISO100/200) or the other efficiency stuff....
My numbers so far are matching closely to what Gabor got though....
And those numbers are all looking really weird....
it's late i will see what i get tomorrow or actually maybe the day after since hopefully down to the shore....
Canon 7D Saturation 13580 @ ISO 100, 15300 @ 200-12800
ISO Value Noise-to-sat Std Dev
ISO 100 0.044-0.045 5.10-5.20
ISO 200 0.041 5.45-5.50
ISO 400 0.052-0.054 6.94-7.15
ISO 800 0.076-0.078 10.1-10.4
ISO 1600 0.123-0.129 16.3-17.1
ISO 3200 0.214-0.235 28.4-31.2
ISO 6400 0.438-0.479 58.0-63.5
ISO 12800 0.868-0.928 115-123
reason wrote:
I'd love to see it too. So far this 7D has most of the specifications from 60D. Not sure why they decided to change their numbering system. Doesn't make any sense! If you don't follow Canon's line it's maybe very confusing for those who just want to buy a camera and need a few choices.
As for ISO noise, they didn't go for 25K so maybe it was all left at the level of 50D. 50D noise performance was a failure and at ISO 1600-3200 didn't really deliver any better performance than my good ol' 40D.
yeah
I hear you, the naming really screw up the system
But, if you look closely the specs is very close to the 1 series but priced ridiculously low.
I am all for it, absolutely.
Just wish they have some nice wide angle lens to go with it. Something that gives 24mm f1.4 kinda thing.
Canon 7D Saturation 13580 @ ISO 100, 15300 @ 200-12800
ISO Value Noise-to-sat Std Dev
ISO 100 0.044-0.045 5.10-5.20
ISO 200 0.041 5.45-5.50
ISO 400 0.052-0.054 6.94-7.15
ISO 800 0.076-0.078 10.1-10.4
ISO 1600 0.123-0.129 16.3-17.1
ISO 3200 0.214-0.235 28.4-31.2
ISO 6400 0.438-0.479 58.0-63.5
ISO 12800 0.868-0.928 115-123
Thanks!
My noise values were taken from rectangular crops on the left hand side outside the image.
I did a visual view of the masked pixels in IRIS, and noticed two things:
1) Above the image, there seemed to be a very thin and not so long crack in the mask. If the Sigma calculations include that area, some weird things are gonna happen.
2) There was a very slight vertical banding in the noise. This will favor vertical crops, but only av very little. I should have done square crops, though.
After looking at Gabor's numbers, I have decided NOT to get the 7D. One of the incentives for me to get a 7D is improved dynamic range at low ISO. If all the 7D can achieve is a mere 11.2 stops (far from the 12 stops attained by D5000). then the deal is off. Even if remove the black current offset from Gabor's results (supposedly the trick employed by Nikon to improve on their dynamic range), I can only attain 11.5 stops.
Looks like Canon is now falling behind Nikon's sensor technology. Not as bad as the pre-D3/D300 era when Nikon was trailing far behind Canon, but certainly VERY disappointing especially after Canon's claim that they aim to improve dynamic range and high ISO performance.
Admittedly, 11 stops is enough for most of my photography needs. But if I want to fork out some dough for a new camera, it better demonstrate its worthiness.
Gabor's numbers still put the 7D at least as good as, and in some cases better than, the 40D (which as its users know well, has very good Real World DR which can be maximised very effectively with smart conversion and processing: even Gabor is a fan of the 40D) so I'm seeing nothing but good news in the figures so far, regardless of source.
I think it's pretty naive to expect hugely improved DR from a camera with these specs, to be honest. I know that opens the door to perfectly reaonable arguments in favour of a smaller sensor, but the 7D is what it is, and it looks pretty good across the board from what I've seen so far.
I use my 40D in some pretty wide DR situations - sunlit white waders/shore birds on black (coal dust) beaches being a common scenario (come to think of it, black and white shore birds against white sand too) - and I've yet to be beaten for DR: the 7D looks like it will do at least as well, so bring it on.
Greg Schneider wrote:
I haven't found any detailed numbers yet...are there any official measures of how many fractions of a stop better the noise is than the 50D? It certainly looks somewhat improved.
Good question...
I would like to see the numbers for RAW....Not JPEG. That is the true measure of the new sensor.
Assuming acceptance of DxO's 40D figures the 7D is ahead of the 40D by a worthwhile amount at 3200, and slightly better at 400 (where my 40D lives) 800 and 1600 (where it often visits).
One should remember that noise varies with spatial frequency (is larger on finer scales); the noise level determines DR; therefore DR varies with spatial frequency. Smaller pixels sample higher spatial frequencies, therefore pixel DR goes down with pixel pitch. Image DR however can remain the same if read noise scales linearly with pixel pitch. Rescaled properly, 40D/50D/7D DR is about the same.
To me, there is not much to choose from in the uncorrelated component of read noise for these three cameras. The big advance in the samples I've seen is the vastly smaller pattern noise. It's about time, this issue has plagued Canon DSLR's for years.
Sure thing, Emil - the lack of pattern noise was one if the things that jumped out at me, and - frankly - noise per se won't be an issue if the pattern noise has gone away, especially as the smaller pixel pitch of this sensor will make the grain of Luma smaller and less noticeable (much as has been observed comparing 40D files to, say, the output of the 30D).
On the DR point, I remain of the view that what the 40D delivers in DR is plenty for the majority of us (it's plenty for me anyway, and I routinely shoot in properly challenging DR situations) and with proper exposure and intelligent conversion and PP decisions, I never bin shots for unrecoverable blown highlights or blocked shadows.
Someone please correct me if I'm wrong, but doesn't read noise have to decrease by the square root of the [ratio of pixel count difference] if the picture impact of read noise is to be kept the same?
In this case - in the 7D/40D comparison - that is sqrt(18MP/10MP) = 1.34 times lower. Or in inverse, the 7D RN should be 1/1.34=74% of the 40D RN.
Otherwise as I'm seeing it the adding/averaging of noise will result in a higher noise effect (both images scaled to an identical, lower resolution), all else kept the same.
Assuming the FWC/amp noise estimations are correct, then the biggest performance increase in the 7D actual picture results should be in the middle-greys and up on low-medium ISO amplifications. For most normal shooters, this is very good.
If AF works (well) and the rest of the camera is mechanically good, then Canon has a winner for the first time in a while! Now if Canon could only get some real colour separation working as well... :-)
]