jorkata wrote:
Deal.
From now on I'm going to call it better SNR for cleaner images at any ISO.
Sorry, still inaccurate. SNR is scale dependent, because noise is scale dependent. SNR at any fixed scale is independent of pixel size, so long as the sensor resolves that scale. What you seem to be vexed about is SNR at image scales that a larger pixel sensor can't generate. I'm not clear on why that is a problem.
ejmartin wrote:
SNR is scale dependent, because noise is scale dependent.
Yes.
Noise tests at 100% magnification tell you about the noise per megapixel count. The 'noise percentage', sort of.
This is a good measurement when cropping, for example.
The noise percentage does not change as you crop.
So, it can tell you, for example, how much noise will be there in a 7D image cropped to 40D dimensions.
I agree that the noise percentage does not accurately reflect the apparent noise at different print sizes.
To resolve this, all that‘s needed is to devise a formula that translates noise percentages into noise per print size and/or ppi.
With this formula, we’ll be able to more accurately estimate and compare noise between different cameras.
You realize that such formula will ultimately be based on pixel-level noise, though, right?
jorkata wrote:
Yes.
And noise tests at 100% magnification tell you the percentage of noise per megapixel count.
This is a good measurement when cropping, for example.
The noise percentage does not change as you crop.
So, it can tell you, for example, how much noise will be there in a 7D image cropped to 40D dimensions.
The same at the same image scale as the 40D, if the 40D is cropped to the same angle of view.
To resolve this, all that‘s needed is to devise a formula that translates noise percentages into noise per print size and/or ppi.
With this formula we’ll be able to more accurately estimate noise.
You realize that such formula will ultimately be based on pixel-level noise, though, right?
The appropriate characterization of noise is its power spectrum as a function of line pairs per picture height. The only way that the pixel size enters this characterization comes from the cutoff on the spectrum at the Nyquist frequency, which will be lower for a larger pixel camera. The noise power at any given frequency in lph will be largely independent of pixel size.
As I mentioned, I don't understand the moaning about noise in images that larger pixel cameras can't generate, while the image that a larger pixel camera can generate, can also be generated from the smaller pixel camera, if desired, with the same SNR.
To me, the issue is what is the best tool for a given use. The finer pixel cam provides the same image information with comparable SNR at image scales the smaller pixel cam can generate, and additional information at image scales the larger pixel cam cannot generate. The same information (and IQ) where potential uses overlap, and more information for a wider array of potential uses.
A couple pages back someone was asking about better reach with a 7D over a 40D. I took some test shots today which may or may not be useful for someone. I didn't have a duck, so I traded that for a box of cereal, strategicly placed in the woods in fairly gloomy light. I didn't mess with any PP, these are straight out of the camera aside from renaming them to make my typing easier. Perhaps someone will wish to try some comparisons with these and make some sort of conclusion. I don't have a 400 mm lens, so I used a 300 F4 with and without a TC. Feel free to DL them and PP them as you see fit. If you want to repost the results in this thread, feel free to do so. (Hoping someone will) I honestly don't get a lot of this technical stuff, never studied it much. For those that do and want to play around with them, here they are.
As you can imagine, these files are pretty large, especially the RAW files. The JPG/RAW versions are the same image, just different formats. Camera was set on low noise reduction which should only affect the jpg's. (I think? Rarely shoot raw.) On a tripod and used the 2 second delay for the 800 ISO shots with the relatively low SS.
brainiac wrote:
Do you get the same result if you (a) process appropriately in DPP, and (b) RESIZE THE SWATCHES TO EQUAL MAGNIFICATION?
200% for each file is not a level playing field. I thought you knew that by now.
That is irrelevant in this case. If a particular unit or run of units is bum and will end up performing worse it is not unreasonable to look into it. I was merely showing that my other canons have not suffered from this issue.
Also, while it doesn't make sense to complain about noise when viewing at 100%, that is not the only thing to look at. Quite a few 5D2/20D/40D users are interested in the 7D in large part for extra reach and/or speed and in the reach case it is perfectly reasonable to care about artifacts and defects that may show up in this manner.
And my 5D2 has more MP anyway, if you want to talk about per frame.
DPP does far, far less of that.
But who wants to be stuck with only using DPP
Later ACR release never fixed maze issues before.
That said... perhaps this time will be different?:
I wonder if this means that it needs a different sort of converter written for it and the standard three primary version won't work right anyway. Perhaps this time the final 7D ACR WILL fix the maze pattern??
Some copies do still band at low ISO somewhat badly, but at least that is only in deep shadows. The maze stuff shows up at regular levels. "
are you kidding?
the 7D is close or even exceeds the 5D2 at high ISO per frame??
for all the talk about 5D2/7D banding the 5D2 appears to do so much better there that it almost never becomes relevant, at least that is my early opinion. My ISO stuff holds up much better so far. Now maybe the occasional picture with poor lighting to extreme and everything down low would cause enough 5d2 banding to make it more bothersome but I think that won't be too often.
actually i partly take it back, if you are shooting with much of the frame barely above read noise levels, in that sort of scenario the banding on the 5D2 can get nasty and then the 7D actually can outdo it at 6400.
ejmartin wrote:
The 10MP mRAW images are nothing more than the camera downsampling the 18MP to 10MP. This is about the crudest form of NR one can do, as it throws out image detail at the same rate it throws out noise. Better is to use one of the good NR filters (NI, NN, etc) to filter out the noise while retaining more detail.
one plus though is mRAW at least bumps up the shot buffer quite a lot which might well make it by far the preferred option for low-light sports photography.
not saying mRAW is the best they could have done, but in the above scenario it may be a better compromise than shooting RAW and reducing/NR later and being constricted in buffer size.
theSuede wrote:
About the latest examples shown by everyone (thank you for your work btw!) I have to add a little something for the LR/ACR-users: Until Adobe fixes the profile for the 7D, it will seem a little bit more noisy than it "should" when used with those converters.
One of the green channels is more sensitive than the other (it has a higher spectral transmissivity - it lets more wavelengths through). It's also expanded towards the red channel by quite a lot.
So to get maximum colour accuracy, you need to demosaic the 7D as "four different colours", it's not R+G+G+B.... It's more like Red+Yellowgreen+Green+Blue. I'm not sure that LR/ACR can do this unless specifically told that the camera needs another demosaic scheme (like the Fuji's and some others)
I've added an example of how this effects colours that are somewhere between yellowish green and warm red. This is the EXACT same raw, with EXACTLY the same settings of sharpening, converted in EXACTLY the same raw engine (my own MatLab development). The only thing that differs is that one is de-mosaiced with an algorithm that supposes that all four colours are totally different from eachother, normal demosaicing usually supposes that the two greens in each RGGB-group are the same colour.
and perhaps this time the final ACR actually will make a difference for the maze stuff and the problem will go away
are you sure the CFA use 4 filter types though and your 4 color thing is simply not just dealing with the unbalanced, but identical greens?
otoh, i also notice it with colors far from warm red and yellowish green, i can see it on white backgrounds, blue, etc. (granted many blues have lots of green in them and of course white has lots too)
It seems like a pretty radical decision if true too.... does it really even make sense for them to do that?
Yakim Peled wrote:
I don't get it. You post actual pictures? But who cares? We want pixels, facts, graphs and math, not pictures. What's the point of using the camera this way?
Happy shooting,
Yakim.
+1
I'm surprised you even unboxed it if all you are going to do is take pictures with it.
One of my friends still has his 20D boxed in the closet.
THAT makes sense.
He didn't want to make charts and graphs so he just tossed it in the closet and lets it sit. You should do the same.
I feel like a fool. After reading all of this, well the two last posts, I realize I have used my 7D and did not need to. If I shoot hi ISO shoots and then study how to make graphs and charts can I be redeemed if I also rebox my 7D?
Ok, seriously, oh, I am serious. lol
I realize that after shooting the 7D yesterday in combination with my 5D and 5D II at an event (a local tribe), I will use the 7D more as a specialty body but not for my main body of work. It is fine and I do prefer the focus point layout over either of the 5D's, by a wide margin, but the FF works the best for me. For my natural history and wildlife work that couples with my visual anthropology, the 7D will be a nice addition.
Nice shots you posted there. I think the camera will come through as a winner. The one thing I did do was video a drum ceremony and the sound and image came through excellent. I also like the way it works a bit more than the 5D II but I still have to get used to it and I don't have any software for video unless it came with the camera but as you can tell, I haven't investigated that at this time.
ExxWhy wrote:
A couple pages back someone was asking about better reach with a 7D over a 40D.
Thanks ExxWhy. I was one of those requesters and this is the kind of comparison I was looking for. Since I shoot cereal boxes about as often as I shoot ducks, these work for me. I'll mess with the RAWs later but I downloaded the jpgs and looked at them both full frame and at 100%. Full frame I don't see much difference between camera or ISO levels other than some exposure and focus variation. This is about what I would expect. At 100% the noise levels look surprisingly similar between the two cameras with the expected increase in noise running from 800 to 1600. What jumped out at me though was that at 100% the noise in the blue part of the larger 7D cereal box looked larger than the noise in blue part of the 40D image. Like the 40D image had been scaled up to 7D size. Whatever the noise levels would have been, since I'm looking at the same number of pixels, I would have expected that the noise grain size to appear the same on my monitor.
I really appreciate your providing these images. I look forward to working with and comparing the RAW files. Cheers! -Jeff
Here's a shot with the combo 7D + 300 F2.8L IS + 1.4x at ISO 800.
I've also added a 100% crop that shows that the sensor provides a lot of detail to work with.
At present, the only way to process the RAW is through DPP, which is a bit of a pain. I worked with RAW+JPG in LR though and it showed noise but it was easy to remove it with NIK's DFIne in PS.
-mirek
jorkata wrote:
With larger wells, S/N ratio is better as well - at the pixel level. Hopefully you are not going to argue about that.
Not interested and don't care about image level comparisons at all.
per pixel yes but...
Generally overall image SNR and DR is the same or slightly better with more MP (do your research) and then you ultimately have no advantage whatsoever with larger wells and less megapixels when the sensor size is the same. Your penalty then becomes less resolution and image detail.
So what's your advantage with a 12mp 7D? There is none.
at least we know not to bother wasting anytime processing RAW files yet (unless you are a DPP user), assuming the different green filters proves to be true, until new versions of third party converters are released. and guess what.... that ultimately leaves MORE time for taking pictures than struggling to get decent processing or having to re-process everything again.
theSuede wrote:
And then, as per request, the scatterplot of the (Green1-Green2) vs Red/GreenAverage saturation following Emils suggestion of a simple
[ (Gr1-Gr2)/GrAverage vs (Gi-R)*(Gi+R) ] to get the correlation between Red channel saturation compared to average green and green channel difference signal.
Same for blue channel saturation.
There's a strong correlation, the only thing messing things up is the fact that the red channel swings back in efficiency as you get colser to the other end of the spectrum, near blue-violet. Therefore, the blue channel (which is a single peak, no swingbacks or plateaus) shows a much stronger correlation. In fact, a VERY strong correlation.
So, there's very strong evidence that:
1) the two green channels are not the same efficiency (one has a "diluted" colour filter) and:
2) one green channel is significantly more yellow than the other.
Not of interest to most people I guess, but some of us really want to know the "why" in some of these things......Show more →
Could it simply be that the sample you tested had the CFA misaligned with the pixel array, so that in one column the nominal G photosite was beneath a mix of the R and G filters, and in the next column the G photosite was beneath a mix of the B and G filters?
The Imaging-Resource test shots include one of a color checker chart; that (preproduction) camera's files also exhibit maze artifacts upon conversion. I'm wondering, if you run the IR colorchecker/resolution chart shots through the same analysis, do they exhibit exactly the same spectral properties for the two green channels?
are you kidding?
the 7D is close or even exceeds the 5D2 at high ISO per frame??
for all the talk about 5D2/7D banding the 5D2 appears to do so much better there that it almost never becomes relevant, at least that is my early opinion. My ISO stuff holds up much better so far. Now maybe the occasional picture with poor lighting to extreme and everything down low would cause enough 5d2 banding to make it more bothersome but I think that won't be too often.
I shoot a 5D Mk II (bought Dec 2008) and also now a 7D. I haven't done any serious comparison, if I HAD to guess, I would say maybe the 5D Mk II has a half-stop edge in high ISO. Like I said, I haven't paid much attention.
The 7D produces images that have a different feel to them than my other Canon bodies. Can't put my finger on it (yet) but it's a very nice effect. I shoot RAW only.
