Canon EOS 7D Master thread

FM Forum Rules
Canon SLRs, primes, and zooms lenses reviews

FM Forums \| Canon-mount SLRs \| Join Upload & Sell	Search Used
1 2 3 … 125 126 127 … 221 222 end

Archive 2009 · Canon EOS 7D Master thread

garyvot
Offline
Upload & Sell: On

p.126 #1 · Canon EOS 7D Master thread

mfurman wrote:
garyvot:
ISO 800 (in bad light), DPP, all NR at 0, 7D (left) vs. 50D (right):

Just to confirm, is 50D image on the right?
To my eye, noise in the image on the left is less but I prefer the image on the right (seems to be sharper)

Well, the luminance noise pattern is sharper for sure; that's one of the major differences in character between these cameras.

This is pure speculation, but from looking at higher ISO images at 100% I believe with the 7D that Canon is taking a page from Nikon's book and is applying some technique at the de-Bayer stage to reduce luminance channel noise. The 7D grain pattern is slightly softened and rounded in a way that Canon's other cameras are not. This seems to allow the files take NR well without loosing too much image detail.

Believe me, I have extensive experience in (trying to) use the 50D at higher ISOs. Above ISO 1250 or so, the luminance noise becomes so coarse, particularly in the shadows, that all fine detail starts to break up and the required NR techniques become more and more intrusive. At higher ISOs, the 50D can make usable images only for use at reduced magnification.

Owning both, there is NO WAY I would prefer the camera on the right to the camera on the left for high ISO work.

Oct 11, 2009 at 01:23 PM

garyvot
Offline
Upload & Sell: On

p.126 #2 · Canon EOS 7D Master thread

mfurman wrote:
garyvot:
If you are cropping for "reach" there is no contest.

I have no doubt that 7D is a better camera than 40D!
I am still not certain that 10 Mpixels out of 18 mpixels of 7D would be better than 10 Mpixels of 40D image (giving a pretty nice TC effect)

Well, in your "reach" scenario you don't need to do any actual cropping to judge noise; just download the RAW files from ImagingResource and compare them at 100% on screen.

Do remember, however, that by cropping any image you are effectively "sub-sampling" the maximum resolution of the lens in use, and any other lens aberations will be magnified. In most cases using a longer lens on the lower-resolution body will result in better IQ.

Oct 11, 2009 at 01:44 PM

mfurman
Offline
Upload & Sell: Off

p.126 #3 · Canon EOS 7D Master thread

brainiac:
You don't seem to be thinking clearly: if you need to crop 10/18 of the 7D frame, then you also need to crop 10/18 of the 40D frame, given any particular lens and shooting distance.

I do not mean to prolong this discussion, taking too much time from you and others but I have to clarify the lack of my "clear thinking".

The comparison that I am trying to make is nothing else than comparing "pixel noise". I do know that this term does not really exist for you because you want to look at the whole image and capabilities of a camera in this respect.

I wanted to look at the same number of pixels images to see if 10 Mpixel images taken out of 18 Mpixels will give a better image than 10 Mpixels of 40D. If that is the case, everything is completely clear to me as far as the progress Canon has made. If not, I will have to find out what size of 7D full image crop is equal to 40D's image.

If I know that, I would be able to see if the increase of "cropability" is worth $1700 for me. I know, as I have already said, that you do not want to think about "pixel quality" and that is the closest, I can come to in my quest to understanding how good 7D is (for me).

Edited on Oct 11, 2009 at 02:04 PM · View previous versions

Oct 11, 2009 at 01:57 PM

mfurman
Offline
Upload & Sell: Off

p.126 #4 · Canon EOS 7D Master thread

garyvot:
just download the RAW files from ImagingResource and compare them at 100% on screen.

I know but is it allowed to look at images 100% on the screen?

I thought that this is the first deadly sin in certain circles. What I mean to say is that there is some justification to compare cameras on a "pixel level". It all depends what one is buying higher pixel density camera for.

Oct 11, 2009 at 02:01 PM

teppy1
Offline
Upload & Sell: On

p.126 #5 · Canon EOS 7D Master thread

in the comparison that garyvot made above of the 7D compared to the 50D at iso 800, i could not see the noise difference until i looked at the cropped versions below. the 7D is definitely better as far as noise. What i did notice though about the uncropped versions was the better shadow detail of the 7D. i can see it in the paper tray and where the ceiling slopes above.

my own observations in shooting HS football at night with each of these cameras, is that the 50D can blow out highlights easily, whereas the 7D retains detail in highlight areas and shadows. The image quality and colors are better in my opinion. I do not ever let the camera do my sharpening, coloring, or anything. I have always preferred to only count on a camera to properly expose and focus. i shoot 'neutral' and take control of color saturation and sharpness in post processing, so there were no in camera settings that make one better or the other for me.

Oct 11, 2009 at 02:08 PM

alundeb
Online
Upload & Sell: Off

p.126 #6 · Canon EOS 7D Master thread

garyvot wrote:
Do remember, however, that by cropping any image you are effectively "sub-sampling" the maximum resolution of the lens in use, and any other lens aberations will be magnified. In most cases using a longer lens on the lower-resolution body will result in better IQ.

Indeed.

However, a very few lenses are so insanely well designed for digital zoom. The 300/2.8 IS is effectively a 300/2.8 - 600/5.6 - 1200/11, not too shabby at the long end:

http://www.fredmiranda.com/forum/topic/820707/3#7620557

Oct 11, 2009 at 02:08 PM

garyvot
Offline
Upload & Sell: On

p.126 #7 · Canon EOS 7D Master thread

mfurman wrote:
garyvot:
just download the RAW files from ImagingResource and compare them at 100% on screen.

I know but is it allowed to look at images 100% on the screen?

Oct 11, 2009 at 02:15 PM

garyvot
Offline
Upload & Sell: On

p.126 #8 · Canon EOS 7D Master thread

teppy1 wrote:
my own observations in shooting HS football at night with each of these cameras, is that the 50D can blow out highlights easily, whereas the 7D retains detail in highlight areas and shadows. The image quality and colors are better in my opinion.

+1. In my limited experience, I've found the 7D behaves more like a 5D/5D2 in this regard (meaning, it's good) than it does a 50D. Highlight detail retention seems quite good and color accuracy is much improved over the 50D in poor light.

What I didn't show in the comparison above is what I had to do to equalize the 50D colors for the scene. Using the exact same in-camera WB setting (3000 K), the 50D image needed a +2 Color Tone correction in DPP to restore proper green channel balance, whereas the 7D color balance is correct for the scene as shot. I shoot in warm light a lot, and find this to be a common issue with the 50D; I'm kind of surprised it doesn't get more discussion here. (My 50D has the latest firmware with the "magenta fix"; this issue doesn't show up so much in daylight-balance scenes, mostly in warm light.)

Here is the image again, first corrected, then at the 50D default setting (7D image can be seen in the top thumbnail):

This image is copyrighted by the owner

Oct 11, 2009 at 02:37 PM

brainiac
Offline
Dedicated FM
Account Locked

p.126 #9 · Canon EOS 7D Master thread

mfurman wrote:
garyvot:
just download the RAW files from ImagingResource and compare them at 100% on screen.

I know but is it allowed to look at images 100% on the screen?

Oct 11, 2009 at 02:52 PM

ejmartin
Offline
Upload & Sell: Off

p.126 #10 · Canon EOS 7D Master thread

mfurman wrote:

The comparison that I am trying to make is nothing else than comparing "pixel noise". I do know that this term does not really exist for you because you want to look at the whole image and capabilities of a camera in this respect.

If you want to compare cameras, then you should look at the same image from each one. It makes little sense to compare different fractions of the image. If you're saying you want to use a higher MP camera for its greater "reach" (greater ability to crop while retaining sufficient detail), then fine, the higher MP camera can do things the lower MP camera cannot. It is not the case that the lower MP camera can do things the higher MP cannot.

I wanted to look at the same number of pixels images to see if 10 Mpixel images taken out of 18 Mpixels will give a better image than 10 Mpixels of 40D. If that is the case, everything is completely clear to me as far as the progress Canon has made. If not, I will have to find out what size of 7D full image crop is equal to 40D's image.

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.

However, if 10MP is what your heart is set on, the 7D should provide a cleaner 10MP image than the 40D -- the 40D at the pixel level will have more demosaic artifacts and the effect of the AA filter will cause the image to be somewhat softer than the downsampled 18MP image, whose demosaic artifacts and AA filter blurring are at scales that will be smoothed away by the downsampling. As far as noise, the cameras are collecting about the same number of photons per unit area, and read noise per unit area seems to be about the same (though as has been noted by many, pattern noise is much reduced in the 7D). Thus, when resampled to 10MP, noise will be about the same (unless one considers digital artifacts of demosaicing error a form of "noise", in which case the 7D will have less for a good converter)

Edited on Oct 11, 2009 at 03:10 PM · View previous versions

Oct 11, 2009 at 03:04 PM

mfurman
Offline
Upload & Sell: Off

p.126 #11 · Canon EOS 7D Master thread

brainiac:
...but aren't you getting completely different angles of view

Do you mean when cropping 18 Mpixels image? Yes, I will but it is the main reason for me to have more pixels. Remember, I do not need more pixels for better quality but "more reach" at the same quality.

Edited on Oct 11, 2009 at 03:06 PM · View previous versions

Oct 11, 2009 at 03:05 PM

theSuede
Offline
Upload & Sell: Off

p.126 #12 · Canon EOS 7D Master thread

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.

This image is copyrighted by the owner

Oct 11, 2009 at 03:06 PM

stiksandstones
Offline
Upload & Sell: On

p.126 #13 · Canon EOS 7D Master thread

I hope ACR gets that update soon, I shot mountain bikes all day yesterday with 7D and converting them in the beta ACR makes me want to sell the 7d. Then I drop in DPP and they are much nicer, which has been said a million times, but I just cringe modifying my workflow in DPP.

Pointless post, but I had to vent.

Oct 11, 2009 at 03:12 PM

mfurman
Offline
Upload & Sell: Off

p.126 #14 · Canon EOS 7D Master thread

theSuede:
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.

Interesting. Does Canon document it anywhere?
If we are going to include the type of RAW converter used, for any S/N or colour accuracy comparisons - we will never agree on anything. I noticed that yellow looked different in some 7D images (comparing with other cameras) but I thought that it was just different light.
I can already here people saying the the reviewers should use different converters to compare cameras.

Oct 11, 2009 at 03:13 PM

ejmartin
Offline
Upload & Sell: Off

p.126 #15 · Canon EOS 7D Master thread

theSuede wrote:

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.

What is the evidence for this (rather than, say, mismatched column gains on particular copies)? Do you for example see a difference between the spectral response of the two green channels in the IR test files (eg the one that includes a colorchecker chart)?

Edited on Oct 11, 2009 at 03:35 PM · View previous versions

Oct 11, 2009 at 03:16 PM

garyvot
Offline
Upload & Sell: On

p.126 #16 · Canon EOS 7D Master thread

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.

theSuede: thank you, this is very good information.

Oct 11, 2009 at 03:20 PM

garyvot
Offline
Upload & Sell: On

p.126 #17 · Canon EOS 7D Master thread

mfurman wrote:
theSuede:
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.

Interesting. Does Canon document it anywhere?
If we are going to include the type of RAW converter used, for any S/N or colour accuracy comparisons - we will never agree on anything. I noticed that yellow looked different in some 7D images (comparing with other cameras) but I thought that it was just different light.
I can already here people saying the the reviewers should use different converters to compare cameras.

Valid point, and one needs to know one's priorities.

I'm a firm believer that to judge a camera fairly one needs to at least include the camera vendor's own processing in the comparison mix. The onus is really on the 3rd party to match its software decoding to the vendor's hardware characteristics.

That said, it's valid to have a greater loyalty to a particular RAW workflow practice than to a particular camera model or brand I suppose. If I could never imagine using anything but Lightroom or ACR and some particular camera looks better or worse there, I would prefer the better one.

Oct 11, 2009 at 03:28 PM

theSuede
Offline
Upload & Sell: Off

p.126 #18 · Canon EOS 7D Master thread

ejmartin wrote:
theSuede wrote:

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.

What is the evidence for this (rather than, say, mismatched column gains on particular copies)? Do you for example see a difference between the spectral response of the two green channels in the IR test files (eg the one that includes a colorchecker chart)?

Such a demosaic scheme gives up a huge amount in resolution -- if all four subarrays are assumed independent, then the resolution drops to half of Nyquist rather than the .7 of Nyquist that would pertain if the two Green subarrays are identical in response and treated as such.

I'm aware of this, and I seen raw-converters that DON'T lower resolution. The greens are sufficiently close together for the engine to work with chrominance-based luma correction per channel. This is quite easy to do.

And yes, I see a different spectral response. Since I'm not to be back at work until next week, I cant give you the exact response curves per channel, but this is the data from an X-Rite 24CC. Channels were extracted as "single channels", and were treated with matched blackpoints and amplification. This meant an offset of 2047 for green1 and 2049 for green2, and an amplification difference of 1.5% (x1.0 green1 and x1.015 green2).

What can't be sen in this chart is that green2 has a different Sdev at the same ADU-level, for roughly 50% ADU I got an Sdev of 78 for green1, and 61 for green2 - implicating a totally different sensitivity. The yellowish green channel has a 1.6x lower amplification per photon to reach the same ADU.

What's interesting in this chart is that all warmer hues have a negative coefficent and all colder hues have a positive coefficent when comparing the two green channel. This does with a very good certainty implicate different spectral sensitivities. The percentages show difference between channels as expressed in percent difference from green1 to green2

This image is copyrighted by the owner

Oct 11, 2009 at 03:37 PM

theSuede
Offline
Upload & Sell: Off

p.126 #19 · Canon EOS 7D Master thread

Oops... never mind the Sdev figures, they were taken on samples that I'd scaled 1:2. Theyre not accurate.

Oct 11, 2009 at 03:57 PM

ejmartin
Offline
Upload & Sell: Off

p.126 #20 · Canon EOS 7D Master thread

theSuede wrote:
ejmartin wrote:
theSuede wrote:

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.

What is the evidence for this (rather than, say, mismatched column gains on particular copies)? Do you for example see a difference between the spectral response of the two green channels in the IR test files (eg the one that includes a colorchecker chart)?

Such a demosaic scheme gives up a huge amount in resolution -- if all four subarrays are assumed independent, then the resolution drops to half of Nyquist rather than the .7 of Nyquist that would pertain if the two Green subarrays are identical in response and treated as such.

I'm aware of this, and I seen raw-converters that DON'T lower resolution. The greens are sufficiently close together for the engine to work with chrominance-based luma correction per channel. This is quite easy to do.

Yes, I realized this not long after posting.

And yes, I see a different spectral response. Since I'm not to be back at work until next week, I cant give you the exact response curves per channel, but this is the data from an X-Rite 24CC. Channels were extracted as "single channels", and were treated with matched blackpoints and amplification. This meant an offset of 2047 for green1 and 2049 for green2, and an amplification difference of 1.5% (x1.0 green1 and x1.015 green2).

What can't be sen in this chart is that green2 has a different Sdev at the same ADU-level, for roughly 50% ADU I got an Sdev of 78 for green1, and 61 for green2 - implicating a totally different sensitivity. The yellowish green channel has a 1.6x lower amplification per photon to reach the same ADU.

What's interesting in this chart is that all warmer hues have a negative coefficent and all colder hues have a positive coefficent when comparing the two green channel. This does with a very good certainty implicate different spectral sensitivities. The percentages show difference between channels as expressed in percent difference from green1 to green2

This image is copyrighted by the owner

This is a fascinating discovery!

What would be great is if you could plot the difference in responses against the nominal RGB values of the various squares on the CC chart (you can get those from BruceLindbloom.com). I think you would want to undo the gamma transform of the RGB values to compare to the linear raw data.

What might make the most sense is to plot the observed green channel differences Gr-Gb against say (Gi-R)(Gi+R) and (Gi-B)(Gi+B) ie the saturation levels of the color differences, where Gi is the average of Gr and Gb; perhaps plotting against (R-B)/(R+B) would be useful too. This would show the trend more clearly I think (especially since I don't recall offhand the RGB values of the CC chart).

Oct 11, 2009 at 04:21 PM

1 2 3 … 125 126 127 … 221 222 end
FM Forums \| Canon-mount SLRs \| Join Upload & Sell 1 2 3 … 125 126 127 … 221 222 end