Nikon has had years of practice cooking RAW files from their subpar Sony sensors, so now their kinks are less kinky with practice where as Canon now has to get kinky with the files for them to be cleaner and compete, essentially due to early climax. Let us not forget the D3x and D300 strangness with luminance noise from ISO 400 to ISO 800.
Summary:
Nikon has had time to smooth out their curves from cooking files while Canon is still experimenting.
odd, my tests gave it 8MP normalized DR of 7.0 for 50D and 7.3 for the worst 7D copy I tested and 7.5 for the best and the 50D I tested seemed to be about average. So it's weird that they got the DR for the 50D and 7D to come out to be pretty much the same at ISO12,800.... and the weird downkink they gave the 7D at the middle ISO.
my lower ISO tests for DR were closer, I had 50D peaking at 11.5 and 7D at 11.7
their ISO ratings seems to show it rates things the same way as the 50D, so a direct comparison per ISO seems to be ok.
whatever the story, things to keep in mind the 7D seems to have more of that cyclic vertical banding but a lot less of the more easily noticeable banding than the 50D, especially at high ISO. In fact, an underexposed ISO12,800 image from the 7D looks pretty smooth if noisy even reduced to websize and a 5D2 looks the same only less noisy but definitely with banding and that mess showing while the 7 doesn't.
my first 7D copy had an IRIS StdDev for the blackframe of 5.8, the second copy had 6.14 and the third one I got to take the test myself from 5.46, or about 1/4-1/5 stop DR best vs. worst. At ISO 12,800 it was 5.1, 5.41, 4.75
the first copy had the most cyclic vertical banding
anyway i'm a little surprised they didn't get it testing better than that for DR, at the least, than the 50D beyond the low ISOs, even the worst copy I ever tested from all the sample blackframes will still 1/5 stop better and the best was 1/2 stop better and most about 1/3+ stop better.
I fail to see the "kinks" in the DxO charts... (?)
I see two things: A very low random readnoise (that's why the curve starts flat from low ISO, and starts to asymptote to the theoretically perfect 1:1 model dropoff at ~ISO400 in the DR) and some un-necessarily high amplification noise (SNR @18% falls just a bit quicker than it should). I also see an impressively high area efficiency, but most of the explanation for that can be seen in the colour-resolution - more than 2x lower than the not-very-good D300s means that the CFA is quite diluted. Which also explains the red-green biased high ISO noise.
There's a very small inconsistency in between the 1600 and 3200 mark on the DR chart, but that was perfectly expected. The last "real" amplification is 1600, anything above this is the result of some math in the DIGIC sensors. This "kink" is VERY small, and does not indicate any unreasonable tampering with the raws.
Skibum: I don't know if DxO use frame-subtraction or not. If they don't, low-ISO DR is probably understated in the charts.
theSuede wrote:
I fail to see the "kinks" in the DxO charts... (?)
In the SNR18% comparison between 7D, 50D and D300s here, one sees the 7D data points are distributed in a very unusual way in relation to those from 50D and D300s.
The data trend implies the 7D starts off with lower SNR than the 50D and D300s, but magically picks up towards ISO 1600 and beyond.
Regardless of what the data means, DXOMark has lost its credibility for me.
mfurman wrote:
I returned it a few days after I got it. I am now able to leave it behind
I command you on your self confidence.
It's not self-confidence. It's common sense. And you seem to have deluded yourself into believing that you've left it behind. You're even at DPR trolling about the 7d. Happy trails...
allnak:
You're even at DPR trolling about the 7d. Happy trails...
You do not even understand what "troll" means? I was posting at DPR a lot in the past (starting in 2003). I do not post their much now at all.
I said many times before that I was waiting for DxO tests because I saw so many conflicting reports about 7D. I came back with that post about DxO tests and I am done with crop cameras in general and 7D in particular. No future posts about them here either!
@thw2:
Yes? That's the perfectly normal trendline in a camera with low readnoise, but good efficiency/slightly high amplification noise. I see nothing strange about it. I have yet to find an error within the individual parameter measurements of DxO - what SHOULD be suspect is their weighing parameters when calculating their "sensor mark". That "combination" score almost never makes any sense to me.
So if DxO's "print" measure is an 8 MP average of a sensor and their "screen" measure is the pixel level measure, then am I right to interpret that the 5D2's 31.2 "screen" SNR is equivalent to comparing the SNR of a 1.6 5D2 8 MP crop to the 7D's 31.1 "print" SNR? (At ISO 800.)
theSuede wrote:
@thw2:
Yes? That's the perfectly normal trendline in a camera with low readnoise, but good efficiency/slightly high amplification noise. I see nothing strange about it. I have yet to find an error within the individual parameter measurements of DxO - what SHOULD be suspect is their weighing parameters when calculating their "sensor mark". That "combination" score almost never makes any sense to me.
No, the slope of the SNR line for the 7D is not normal. The normal is 3 dB decrease in SNR for each ISO stop. Amp/conversion noise contributes nothing at 18% gray. Read noise comes in at higher ISO's and makes the curve steeper again. The 40D is very close to 3 dB/stop in the 200-1600 range. The 7D has about 2.5 dB decrease per ISO stop. High ISO perfomance is optimized in the 7D, but that performance does not keep up at lower ISO's.
One possible explanation would be that the photosites in the 7D have a nonlinear response, and are being linearized in hardware/firmware. In that case, the photon noise would behave as the DxO curve shows. Ejmartin's measurement of system gain does however not support the nonlinear response theory.
Other possible explanations are that the noise at lower ISO's is not only photon shot noise, but either individual pixel response variations or column gain fluctuations as well.
theSuede wrote:
@thw2:
Yes? That's the perfectly normal trendline in a camera with low readnoise, but good efficiency/slightly high amplification noise. I see nothing strange about it. I have yet to find an error within the individual parameter measurements of DxO - what SHOULD be suspect is their weighing parameters when calculating their "sensor mark". That "combination" score almost never makes any sense to me.
have you measured 7D DR at all yourself?
i'm just surprised why my measurements are suddenly noticeably different from theirs at the highest isos, i get a reverse trend from what they do actually (while with the 40D/50D I get the same trend they do and pretty much the same for the 5D2 as well)
mfurman wrote:
You did not read their "Insights", did you?
Seriously, should one expect 'insight' from a site whose default method of comparison between 10 and 18 Mpixel APS-C cameras is to compare on a per pixel basis? It's great that they provide the 'print' graphs, but they are not the default, so people like jorkata go around citing the bogus graphs as if they offer any kind of worthwhile insight. DxO is a bad joke, like DPR and LL. It's amazing how hard it is to find technically correct assessments between cameras. Re-sizing imaging resource files is about the best way to get any insight.
There scores dont make any adjustment for resolution or detail do they?
Anyhow, comparing my old 350d to the 7d results shows very little difference between the 2 cameras, which is somewhat odd considering the images they both produce.
theSuede wrote:
I have yet to find an error within the individual parameter measurements of DxO - what SHOULD be suspect is their weighing parameters when calculating their "sensor mark". That "combination" score almost never makes any sense to me.
Actually, I am not thinking of their 'combination' score (which is absolutely garbage anyway).
What I don't understand is their assigned score for individual sensor performance area is also suspect.
Consider this: all their data, i.e., SNR18%, dynamic range, color sensitivity etc (viewed either at the pixel OR image level), indicate the 50D high ISO performance is nearly equal to that of the 7D. YET, they assign the 7D a score of 809 for low light ISO and the 50D only 696.
What’s going on here? Is that because they do not trust their own analysis or their methods are completely arbitrary to begin with?
RobDickinson wrote:
There scores dont make any adjustment for resolution or detail do they?
Anyhow, comparing my old 350d to the 7d results shows very little difference between the 2 cameras, which is somewhat odd considering the images they both produce.
The same can be said for 450D vs 7D. As a matter of fact, the 450D is sharper both on the pixel as well as image level. This is true from f/4 onwards so it cannot be explained by diffraction alone. Weird.
mfurman wrote:
I am wondering what Internet source of proper tests you would recommend.
Like I said, if you resize imaging-resource files to the larger dimensions you can get a fairly good idea of comparisons between cameras. Nothing beats actually trying out two cameras in real working circumstances though, ideally with similar or identical lenses, and equivalent focal lengths.
