Need 50D sample image data

[Roger Clark]

Hello 50D owners. I need a few sample images from a 50D that would take only a few moments of your time. If you are not aware, I do analyses of digital camera sensors along with a few other astronomers and physicists. I maintain a summary page of sensor analyses and the implications:
http://www.clarkvision.com/imagedeta...rmance.summary

As you may see from the above page, we have some data on the 50D sensor from Christian Buil, but it is incomplete. I need a set of images to characterize ISO 100 to 1600. Here is specifically what I need:

With the 50D in a room temperature environment (about 68 to 75 F, and been there for at least 1/2 hour), take pairs of exposures in a dimly lit to relatively dark room, say at night with only an single light on, with the lens cap on, manual mode, f/16 or higher, exposure time: 1/4000 second. Take 2 at each ISO: 100, 200, 400, 800, and 1600 (10 exposures total). Wait about 5 seconds between exposures, and record the data as raw. Then I need one other exposure taken at ISO 100 (any f/stop or exposure) where one element in the scene is obviously over exposed. For example, an outdoor scene with clouds blowing the highlights. That exposure lets me see the maximum signal the sensor will record, and the fast exposures tells me what is known as the read noise.

Then arrange to email me the raw files, or send them via a CD/DVD, or if you have an ftp site, where I could ftp them from.

I'll analyze the data and put the results on the sensor performance web page. You'll receive a credit for supplying the data and have the knowledge of the performance of your specific camera.

If interested, please email me at: rnclark at qwest.net (Please do not just email raw images as it will quickly overload my quota; I'll need to take a few at a time and clear my inbox.)

Thanks.
Roger

[Tim Dodd]

I'm happy to provide such sample files. Email sent. Awaiting reply. :)

I'm sure you won't need them but FWIW I also have a 30D and 40D so could provide equivalent data from those as well..

[Roger Clark]

Hi Tim,
Thanks. I sent you an email.

I also have a 30D and analyzed a friend's 40D, so those are covered in the sensor page. The 20D and 30D have the same sensor. I'm planning on buying a 5DII, so that will get covered. Now if someone has the new Sonly 24 megapixel, or a recent Pentax, that could be interesting.

Roger

[Emil Martinec]

Raw data I have yields for the 50D

• ___ ISO__Rd Ns, ADU___Rd Ns, e-
• ____100_____6.2_______13.7
• ____200_____6.9________7.6
• ____400_____8.5________4.7
• ____800____12.4________3.4
• ___1600____20.6________2.8
• ___3200____41.1________2.8

where I have used Christian Buil's gain value of 2.2 e-/ADU @ ISO100 to infer the read noise values in electrons. The high ISO read noise in electrons is the best per pixel value for Canon DSLR's so far. Unfortunately this is tempered by some substantial banding noise, especially at high ISO, as others have noted. Buil's value for the gain represents an area efficiency for photon capture that is about 5% better than the 40D. Per area read noise values range from about the same at ISO 100-200, to about 20% lower for the 50D at ISO 1600 (an achievement again marred by the substantial banding problems).

[Peter Hawrylyshyn]

roger -
can you just reconfirm:
take pairs of exposures in a dimly lit to relatively dark room, say at night with only an single light on, with the lens cap on

"with the lens cap is ON" - - why does lighting in the room matter??

can the 11th photo be an indoor scene, or must it be outdoors

will gladly help out
peter

[Roger Clark]

Peter,
Yes, I specified with the lens cap on and dimly lit room in case the lense cap is not a perfect seal. We don't want any light on the sensor.

The 11th photo can be anything that saturates the signal, so indoors works too.

But, I got the images needed from Tim Dodd. I'll post results shortly. They are very close to Emil's results.
Thanks for offering to help. Maybe the next camera....

Roger

[Roger Clark]

Hello Emil, Everyone,

Interesting. I got almost identical results from Tim Dodd's camera, results below. I agree that the read noise sets a new standard and is extremely impressive. These results explain the impressive high ISO images that have been posted here at BPN recently. While the banding concerns me for some applications (like astrophotography), it does not seem to have affected the image examples posted so far.
The low read noise you got along with the values from Tim's camera and Christian Buil's result all indicate a consistent picture of low read noise in 50D cameras, and a new and better standard for other cameras to equal. Now to control that banding.

______________________Maximum______________
ISO___Gain___Read Noise__signal____Dynamic range
______e/DN___electrons___electrons____stops

_100___2.2_____13.83____27300______10.95
_200___1.1______7.55____16900______11.14
_400___0.55_____4.60_____8450______10.84
_800___0.27_____3.23_____4200______10.34
1600___0.14_____2.61_____2100_______9.65
3200___0.07_____2.69_____1050_______8.61

AIQ = 52.7

You can see the AIQ values for other cameras in Figure 9 at:
http://www.clarkvision.com/imagedeta...rmance.summary

I'll be adding this new data to my other figures on the above web page. In the meantime, you can see where they plot by locating 4.7 microns on the horizontal axis and plotting a point close to the model line. All the data are close to the models, so performance is close to what I would have predicted. The low end noise is still dominated by the 14-bit A/D converter, although it is clearly better than that in the 40D. I look forward to the next generation that hopefully will have 16-bit converters and even better performance. ;-)

Roger

[John Chardine]

Roger et al.:

1. What is the cause of what I would characterise as positional auto-correlation of noise, otherwise known as "banding".
2. Do all sensors suffer from banding or are some worse than others. Is the 50D particularly bad in this regard?

[Roger Clark]

Roger et al.:

1. What is the cause of what I would characterise as positional auto-correlation of noise, otherwise known as "banding".
2. Do all sensors suffer from banding or are some worse than others. Is the 50D particularly bad in this regard?

Hi John,
"Banding noise" or fixed pattern noise has several causes,including what is called reset noise, uniformity of the sensor array, stability of voltages, and electronics getting the signals off the sensor. All sensors have this noise. A decade or so ago such fixed pattern noise was much worse and one have to do a lot of work to calibrate it out, even in scientific images. We are complaining about a small effect in comparison. Consumer digital cameras have really come a long way. We are just hoping for that perfect sensor! ;-)

You can see banding in images on this page (1D Mark II images): http://www.clarkvision.com/photoinfo...ht.photography
See, for example, the bottom two panels in Figures 5, 10, 11, and 12.

Unless you do astrohotography or other low light imaging where there are large uniform dark areas that need boosting, the banding is probably rarely seen in 50D images (or 40D, 1D Mark III).

I've had frustrating issues with banding on my cameras, like 1D Mark II with night photos, like this one (you can't see it in the web image):
http://www.clarkvision.com/galleries...ax.cw-800.html
so the 50D banding concerns me. Some banding can be calibrated out, like done on the night.and.low.light.photography web page above.

The 40D showed a new standard for smooth low end having the least banding of any camera introduced up to that time (the 1D Mark III is similarly good), so people expected better from the 50D. I think the banding in the 50D is actually quite low, just that it becomes more apparent because the random noise (read noise) is so low.
So it may be that the banding in the 50D is just as low as the 40D, just it is more apparent because the 50D read noise is so much lower (60% lower). I'll have to look at this more.

I do this camera research in the quest for a better astrophotography camera. But then it also helps with high ISO performance for those wildlife action shots near sunrise and sunset.

Roger

[John Chardine]

Thanks Roger. Very interesting. Lots of talk on the web as if the 50D was the first digital camera to experience banding!

[Tim Dodd]

If anyone is interested I have shot a whole bunch of test images with my 30D, 40D and 50D in all variants of raw/sraw formats and at each available full stop iSO. I then processed the files through DPP and Lightoom using default settings for sharpening and noise reduction. I've uploaded the full resolution files as processed JPEGs to an online album here....

http://picasaweb.google.com/EezyTige...ey=7KDmp1s3qLg#

There is a discussion thread I started on DPReview where you can find links to download the full resolution JPEGs near the bottom of page 1 of the thread.....

http://forums.dpreview.com/forums/re...hread=29918648

I have avoided pixel peeping the results as I can't really be bothered to pore studiously over 84 image files at pixel level, but I have offered my opinion regarding the practical useability of the files at various ISOs when viewed as a full screen image on a 17" WUXGA monitor.

[Emil Martinec]

Thanks Roger. Very interesting. Lots of talk on the web as if the 50D was the first digital camera to experience banding!

No, all cameras have it to one degree or another. I've been thinking about a testing procedure to measure it quantitatively. Typically its contribution to the fluctuations of a typical pixel value is a good bit less than the overall fluctuation level, however the eye picks up on it because our vision is adapted to detect lines, edges and patterns.

Some pattern noise is truly fixed and not varying from image to image; that is easy to contend with. Tougher to deal with is pattern noise that varies from image to image, as one can't make a master template to remove it. I did some tests a while back on my 20D that illustrates their relative importance on that camera:

http://theory.uchicago.edu/~ejm/pix/.../#patternnoise

[Grady Weed]

Roger,

My head hurt after reading this thread! Seriously, I did find it interesting on the technical side. This helped me to understand why my Mark 2n has had some issues with low light and the lighthouse, lens not focusing, pixels blown or color spots after long exposures etc. PS: You have one nice web site Roger. Loved the astro images!