Zenon Char wrote:
A few from a football game tonight. ISO 12,800 and cleaned up using Dfine. 300L F4 IS @F4, Shutter speeds averaged at 1/320. Ai servo using center focus point. Oh yes. High speed continuos.
Holy crap those are good - very well done.
I didn't see the EXIF, but are you sure 12800 - sorry, I gotta ask.
Also, must be some crappy light since even at F/4, you can only get 1/320s.
Alek Komarnits wrote:
Holy crap those are good - very well done.
I didn't see the EXIF, but are you sure 12800 - sorry, I gotta ask.
Also, must be some crappy light since even at F/4, you can only get 1/320s.
I use save for web. Just a habit for entering on line contests. File size can't be bigger than 300 so I try to use that for images only. I'll do another with the EXIF for you.
Here you go. I am assuming if I use "Save As" in PS that will keep the EXIF. I don't use that very often. Actually the TV was 1/500 on this one. Most were 1/320.
Here's a shot walking out of the hockey game last night.
Shot in full-res raw, converted to resized jpeg with all default settings... I didn't change anything.
Nothing special obviously, just a snapshot.
The first test gives a wealth of visual information, as you see the difference of neighboring columns and how badly out of balance they are. The second test is a quick test to see whether on average the two green subarrays are differently calibrated.
Thanks Emil, I used Iris to do as you suggested and don't see banding at that level. The histogram also looks exactly like one would expect (without all the spikes).
thedigitalbean wrote:
Thanks Emil, I used Iris to do as you suggested and don't see banding at that level. The histogram also looks exactly like one would expect (without all the spikes).
The example in the first link was generated using ImageJ, another image data analysis program (BTW the spikes in the histogram of the example are due to the way ImageJ bins data and have nothing to do with the RAW data itself). The level and contrast were adjusted to make the results easier to see in the posted screen shot, but not dramatically.
Basically, if you can easily see banding above the usual level of noise then there is cause for concern; otherwise there is nothing to worry about. Note that this test is not the sort of extreme manipulation that people have been doing for offset banding -- you know, push the exposure many stops and see how bad it looks. Here, if you see something in the difference of the green channels in a normal exposure, that means it rises above the level of photon shot noise and will affect the image, at least in terms of noise and perhaps worse effects such as mazing.
Unfortunately most of the tools for this sort of analysis, other than ImageJ which is Java-based, lie in the Windows world. I'm actually a Mac guy; I run them under Crossover.
Here's a shot walking out of the hockey game last night.
Shot in full-res raw, converted to resized jpeg with all default settings... I didn't change anything.
Nothing special obviously, just a snapshot.
Fred Tedsen wrote:
Is there a way to check for problems using a Mac? There should be a way to see problems in images directly if something is obviously wrong, but what conditions will most reliably show them?
Yes, in fact I just thought of a method that should work and is reasonably simple.
2) Take a test image of some tonally smooth object, make it somewhat OOF so that there is no texture in the image itself (for instance a colorchecker chart is a good choice; the different colors enable one to diagnose the problem better, should it occur; for instance, in skibum's bad copy the problem was most pronounced in yellows and not so much in blues).
3)Convert it two different ways using dcraw, once as a three color image and once as a four color image. For instance in a Mac terminal window go to the directory with the test image and execute the two commands
(Notes: I had to convert the test image IMG_5707 to dng because I haven't yet updated to the latest version of dcraw, which now supports the 7D; and obviously, you should substitute the name of your test file. Also, the end of the command " > IMG_5707-4color.tiff " pipes the output to a file of that name; the syntax might be a bit different on windoze, I wouldn't know.)
4) load both images into Photoshop and overlay one on top of the other; set the blending mode to difference. Make a levels adjustment layer and bring the difference image up a few stops to look for any patterns (it's the difference image, so it should have been black if the two conversions agreed with one another). You'll want to try both layer orderings (4-color either above or below 3-color) since a priori one doesn't know which if either might be brighter if there's a channel imbalance.
Here's what I got with skibum's bad copy color checker test image (this is at 200%, and two stops brightening of the difference image, for visualization purposes):
The mazing is directly the result of the green channel imbalance; with 3-color mode, dcraw does its standard interpolation, while with 4-color mode it averages the two green channels from diagonally adjacent pixels before interpolating. Since in the OOF image there is no texture, the difference image shows only artifacts that are generated by the mismatch in greens.
gfiksel wrote:
Just wanted to mention that Lightroom 3 beta is a HUGE improvement as far as the color rendering and noise handling. In addition, it recognizes Canon mRaw and sRaw formats! Hopefully, some of the profiles will get into LR 2 before the LR3 is out, maybe the next update?
The first image resized, ISO 6400 at f/5.6 and 1/50 sec
The second one is 100%, no noise reduction, no sharpening.
The third one is a modest color noise reduction. Very nice, tight grain.
.