Printable View
I've been testing a 7D this weekend
The following images of a piece of birch bark were taken on overcast day about an hour before sunset using a 400mmDO (in manual focus mode) on a tripod wide open at f4 at various ISO settings on the 3 cameras. Shown at 100% crop (250 px by 250px) . All processed at same settings with ACR in PS with no sharpening added.
Objective was to assess comparative image quality/noise - will let the images speak for themselves
Thank You
If possible could you processes these also using the Canon DPP program. From what I have heard it does a better job.
Bill
Hi Peter, Which version of ACR was used to convert the RAW files.
Thanks
Bill - i've never used DPP in my workflow. I've often been unhappy with image noise using the 50D, so I was most interested in how the 7D would compare to the 50D at comparable ISO settings.
Don - i used the latest update available with PS - CS4
Attached is another comparison. IMO the 50D does show much more noise at ISO 400 and 800 when viewed at higher magnifications
Don't want to chuck any cold water on this lads, but assuming Peter has used the same processor for each example, (and I don't see why he shouldn't): the important points remain in that there are clear differences. It doesn't surprise me as there are other forums and tests being done that indicate the same story. Increasing pixels in Canon's case, is not an indicator to getting better results, but heck, who can compete with the marketing gurus where you perceive you'll get better and thus buy the new model. There is a bubble waiting to be burst here, it's just a question of whether its your bucks involved. D
Dave -
In going from 15MP on the 50D to 18MP on the 7D , from what i can see Canon DID reduce the image noise
The $3000 question next month will be - how much better is the new Mk4
Another thing that may help the 7D over the 50D is that the 50 has a Dual DIGIC 4 processor while the 7D has Dual DIGIC 4.
Bill
The 50D images look more noisy, but it look more sharp too against the 7D. Is it the antialysing of the 7D?
best regards
paulo anjo
Paolo -
The 50D and Mk3 had AF Microadjustments to calibrate the AF of the 400mmDO to each camera body
The 7D is a new demo model with no AF micro-adjustments
This may in part account for slight differences in sharpness
I was only Interested in which ACR version was used because there is a noticeable difference in the RAW file conversion from the Beta 7D support and the official 7D support knowing that their is a difference I only wanted to make and informed judgement on what I was seeing.Quote:
Originally Posted by David Fletcher
Hi Peter
I think you have a couple of problems in this comparison.
First, there is movement in at least some of the low ISO samples (I can see slight double edges). Second, you have magnified the images from the 50D & 7D more than those from the 1D3 (which emphasises the noise in the higher pixel cameras). It would be interesting to see some same-size image noise comparisons between these cameras if you have the time/inclination.
John
John
My interest wasn't image sharpness - just to assess the noise in the images. Some of the movement would be due to wind, .... given the shutter speeds.
These are 100% crops (250px by 250px) from each sensor. The differences in image size you see are due to the size of the sensors. Same size images would impact the noise seen due to down-or-up sampling. Having looked at a bunch of other comparative images , the 7D does consistently have less noise than the 50D.
Peter,
Comparisons such as you show have numerous pitfalls. Perception of noise is dependent on intensity in the image and the color of the background and clearly your 50D images are brighter, thus biasing the test toward the 50D. Second, the raw converters may be using different settings for the different cameras, even within the same converter. The raw converter does averaging of the surrounding pixels, for example, and different raw converters will and do produce different results.
The only reliable and unbiased way to determine true noise is to analyze the raw data from the sensor before any raw conversion (and even then the camera may have modified the sensor data--but the raw data are the lowest level we consumers have access to).
I have analyzed numerous sensors and the results are summarized at:
http://www.clarkvision.com/imagedeta...rmance.summary
If you look at Figure 2 for example, you will see that the signal-to-noise ratio on the Canon 1D Mark III, 5D Mark II, 40D,50D and Nikon D300 all closely follow the same modeled line. And older generation cameras fall below the line. The main factor in the trend is pixel active area. The older cameras had lower fill factors (a smaller fraction of a pixel was sensitive to light). Note the older Canon S60 and S70 cameras also fall on the trend line. The S60 and S70 are CCDs and have high fill factors.
The reason for this trend is basic physics. The light hitting the sensor is fixed in a given image, so as you chop the sensor into smaller pixels, each pixel gets less light. Most of noise we see in our images is due to that from photons (noise = square root of the number of photons collected). So the results from any new camera are quite predictable in spite of manufacturer hype.
The Canon 7D has a pixel spacing of 4.3 microns. So you can plot on Figure 2 (and many other figures) where the 7D will most likely be when we get real sensor data. And the prodiction is lower than the 50D because of the smaller pixels.
Where manufactures have really been improving sensors is in the lowest level noise, seen in the shadows at low ISO and all throughout images at high ISO: read noise due to the sensor read electronics, and fixed pattern noise. Unfortunately, analog to digital conversion noise has not improved in the data I've seen, even in the change from 12-bit to 14-bit A/B converters.
If someone would like to take the data for me to analyze the 7D sensor, I would be happy to do it. It would take about 1/2 hour of your time acquiring the data and you could either email it to my or send it via email (I have a special email address that can accept large files; not the one in my profile), or if you have an ftp site, I can pull it. (Maybe I can set up an ftp site too.)
It will take me about 40 hours of work to do the analysis and a little longer to get the data online. I can probably get it done by about New Years.
The offer also holds for other cameras that are not on my sensor performance page. But considering the time to analyze the data, I can't do too many cameras at once. All I ask is once you send me the data that I have permission to freely use the images, which are of a blank sheet of paper and some dark frames. I'll add your name to the credits if you wish.
So just contact me. If you decide to help out, here is what you are in for:
This page shows my general procedures (in case anyone wants to do this themselves):
http://www.clarkvision.com/imagedetail/evaluation-1d2
and here is how to take the data:
http://www.clarkvision.com/imagedeta...otakedata.html
I am interested in the deriving data for the following cameras:
Canon 7D, Canon 1D Mark IV, latest Nikons, latest 24 mpixel Sony, latest pentax, latest micro 4/3.
Roger
Roger -
I was hoping/expecting you'd reply. Mine is the visual world of what i have to correct in PS, yours is the true world of quantum physics. I'll see if i can get you the data for the 7D
Peter
And ironically, the noise we see in our digital camera images is due to quantum physics. So we are all observing quantum physics. (Pretty cool that technology has come this far.)Quote:
Originally Posted by Peter Hawrylyshyn
Roger
Peter -Quote:
Originally Posted by Peter Hawrylyshyn
Although I have yet to use it, my 7D does have AF micro-adjustments: C.Fn III-5
Quote:
Originally Posted by David Fletcher
It is generally true that marketing always pushes for more and most people think more is better but in the case of 7D it really is. I did some controlled tests between 7D and 50D (which I have sold and no longer have) and found that 7D has indeed lower per-pixel noise than 50D despite smaller pixels, Canon data shows improvement is due to new ML design, more transmissive CFA and better cell engineering which leads to increased external cell quantum efficiency. 7D pixel pitch is smaller than 50D but effective cell area is larger thus each cell "sees" more light. Furthermore the AA filter on the 7D is a different design (similar to 5D MKII) and it really does take advantage of all those 18 mpixels *in good light, when AF is spot on and when the lens has enough optical resolution*. Here is a 100% crop of 7D with EF 400 DO wide open (f/4) at ISO 400 with no NR, This is better than anything I got with 50D. Of course the difference between MKIII or FF cameras and 7D in terms of per pixel noise still remains significant, but the 7D has the best IQ of any crop camera and its extra pixels in many (NOT all) cases make up for its higher pixel noise compared to MKIII.
http://www.stanford.edu/%7Eahazeghi/...s/IMG_0924.jpg
The main problem with OP's test is that 7D images are OOF or have shake and thus the detail is lost. If they are all in perfect focus 7D's advantage will be easily seen over the other two cameras. Overall 7D is a very nice camera, its AF in certain situations remains its weakest link IMO and the only reason anyone might pick a 1DMKIII over the 7D.
Doug has some nice crops with 7D and 500 f/4, I am sure he can post some as well. For best results use DPP to convert the RAW files.
As was mentioned, there is a BIG difference between the beta support for the 7D in ACR 5.5 and the near final version in ACR 5.6rc (Release Candidate, usually at this point all they will change is the color profile). The cause for this is a huge amount of column variation in amplification which has to be dealt with before demosaic, otherwise one will get the maze artifacts that are quite apparent in the OP. In addition to this, there seems to be a large sample variation in whether the two green channels in the Bayer array have the same response to different frequencies of light (the fact that it varies from copy to copy suggests that it's poor QC in aligning the CFA with the photosite array during manufacturing). This also causes mazing if not dealt with. None of this will show up in the methodology Roger typically uses for testing read noise and gain, which uses difference images to subtract out pattern noises. To give an idea how big a problem it can be, the worst camera I tested had a std dev of signal in uniform OOF areas that was twice the photon shot noise in well-exposed areas at ISO 100:
http://forums.dpreview.com/forums/re...ssage=33491828
With these flaws in the RAW data of the 7D, it is going to make a much bigger difference than usual what RAW converter is used, and whether the programmers of the converter have installed special code to deal with the 7D (as apparently Adobe has between the beta and final support for the 7D in ACR).
And a little side note for Roger: read noise is quite good on the 7D, as I recall about 2.3 electrons at high ISO. Gain is also quite good, about 2.15 e-/raw level at ISO 100 (see http://forums.dpreview.com/forums/re...ssage=33335705 for a plot), putting it at D3 levels on a per area basis unless Canon has "cheated" by changing the ISO normalization (again). But the pattern noise is a bear.
Just wanted to note in connection with this the following from the-digital-picture:Quote:
Originally Posted by Emil Martinec
"my 7D delivered exposures that are 1/3 stop darker than my compared-to bodies at the identical exposure settings in this test."
http://www.the-digital-picture.com/R...ra-Review.aspx
I hope that the Dxomark results on ISO (and noise) will give a more definitive answer.
Yes, but the raw convertor you used is not a final version for the 7D, it is a RC (Release Candidate) and while the others cameras raw conversion is probably set. To be valid you'll have to do the test all over once the final version of ACR 5.6 is released. Also doing a noise test under windy conditions is a variable that will affect the final outcome. It would be best to do it again under more controlled conditions. It's also necessary to understand that the settings in ACR are meant to be just a starting point and for a "visual world" comparison the tester is should manipulate the settings to get the best image possible. After all that is what I do in the visual world, I don't just accept Adobe's default settings in ACR as the best for my images.Quote:
Originally Posted by Peter Hawrylyshyn
Actually, unless one is interested in understanding how one's raw converter of choice copes with noise, it is best to look in tonally flat, OOF areas. So windy conditions, or less than perfect focus, are not so relevant to looking at noise. Better one set of tests targeted at resolution, and a differently directed set of tests for noise.
Emil,Quote:
Originally Posted by Emil Martinec
To clarify for everyone, the method I use is designed to avoid the fixed pattern noise so one can derive how many photons each pixel is capturing, what the maximum photons per pixel is, and the absolute calibration of camera DN to the number of photons. I did not come up with this method; it was derived by sensor manufacturers. The results tell the fundamental response of the sensor to light.
But since the method takes single frames for many light levels, banding issues and fixed patetern noise is also recorded and can be analyzed.
Thanks for the info. Did you derive a full well value?Quote:
Originally Posted by Emil Martinec
Roger
Indeed you can use the same images to analyze the pattern noise (in this case fixed pattern noise), I simply meant that it might go overlooked if one wasn't aware that it's an issue with this camera. If you end up doing your usual analysis suite, I'd be interested on your take regarding this issue. I spent a lot of time on it, and ended up being thoroughly dismayed at Canon. It appears to be a per column gain variation that could easily be calibrated out in firmware, if they wanted to. In addition to this multiplicative shift, the eight-channel readout gives an eight-column periodic read noise banding that would also be easily calibrated out, if they wanted to. The read noise banding is not a big deal, but the gain banding causes real headaches for RAW converters, as you see with ACR 5.5ß
Unfortunately, the test images I got didn't include clipped highlights, so I don't have a full well figure. I might have some clipped highlights from other RAWs from other cameras, but not from that particular camera that I developed the gain analysis from. The person who supplied the test images has since returned the camera for a replacement, that had the same issues, though to a lesser extent.
The Dxomark data on the 7D are out. The 7D's ISOs are overstated exactly as on the 50D, so e.g. when the camera says you're on ISO 200, both the 7D and 50D really give you ~ISO 160 per Dxomark. At that ISO, the 18% gray S/N (signal to noise ratio) is worse for the 7D by 2.2 decibels. I think :) that means roughly 2/3 of a stop. The 18% gray S/N is a basic measure of noisiness shooting an overall 18% gray subject. At higher ISO's the two cameras' S/N at 18% gray converge. The 7D improves slightly over the 50D on Dxo's measure of "color sensitivity" which I don't understand.
The S/N calculations are based on a similar model to Roger and Emil's that does not take into account any correlation in noise between pixels. Thus it would not account for the observation often made that the 7D's noise is finer, a further advantage for the 7D over the 50D.
EDIT: I somehow misread the graphs. Aaaargh. The 7D is actually worse than the 50D at lower ISO, not better. They converge at higher ISO's. This is with the "Screen" tab in Dxo which I think implies a per-pixel comparison. They also have a "Print" tab which does a per-area comparison. Need to be more careful.
http://www.dxomark.com
Bottom line is the 7d a worthy upgrade from the 40d?
