Separate names with a comma.
Discussion in 'General Technical Discussion' started by Iliah, Apr 13, 2005.
Thanks for the illustration.
I wasn't sure what to make of it at first, but looking more closely, particularly at the User Gamma Curve in addition to the differences in tonal range gradations (and resultant details) that are visible in the rendered images, it looks like you're trying to show us that while the analog "side" of the D2X sensor may have significantly less DR than the 1DsMk2 counterpart, the digital conversion "side" does benefit from not having to squeeze the extra DR into the same bit space, which means more tonal values to spread around for each stop of light capture.
So in the end, there's a subtle, little-spoken-of benefit -- and perhaps, even a wise, intentional compromise on Nikon's part -- that we get from the smaller DR. It's certainly not so obvious as getting "greater" DR and also not as easy to explain to the masses, but it's a fact of life. And perhaps, this is even one thing that contributes a little (or maybe a lot!) to our perception of the plasticky look of Canon(!) now that I think of it.
In order for the camera to truly benefit from the extra DR w/out this kind of compromise, it will need to at least increase the bit depth of the A-to-D conversion (and the RAW file size too) accordingly, but Canon has not done this yet.
BTW, it seems that you even tried to make sure we see the point by rendering the Canon image more contrasty, stretching out whatever tonal values there are in the shadows and midtones so we can see if there are much of any to speak of.
So I guess while we need to be careful not to blow highlights w/ the smaller DR, Canon folks will need to be careful not to underexpose because there's not much details to recover in the shadows.
Hmmm... I wonder if Canon's stretching of tonal values is contant or variable according to meter reading. I imagine it's either constant or doesn't have too many steps if variable. Could have a pretty real negative impact on properly exposed images that don't make use of the wider DR, but need fine details and tonal gradations in the shadows and lower midtones...
Very good and enlightening illustration. Thanks much.
PS: BTW, does this mean RAW Magick will soon support Canon RAW too?
I'm really not sure that I understand what I'm seeing here. I think I understand what _Man_ is saying, but it isn't at all clear to me that this is what you were trying to show.
To my (very, very) untrained eye, the D2X looks better, which surprises me. Maybe I'm just not looking right. :| OTOH, that may be what _Man_ said in much more detail. ;-)
Gotta disagree Frank, to my (equally, if not more so) untrained eye the 1DsII file looks to have better colors at least (look at the bricks) and contrast. However, I strongly suspect Iliah was trying to show something a whole lot more than that and in a less subject way than I've given.
In RAWMagick histogram you see is *after* the curve shown in the same tab is applied.
The point of the post was to show how much more bending of the linear curve in Canon file is needed to get more or less similar histogram.
Shifting dynamic range towards highlights leads to the waist of DR for the vast majority of the shots that will not benefit from that. But shadows are sacrificed, and noise appears.
DR in both cameras is very close to being equal, but centered differently.
Different centering also leads to discussions of underrated vs. overrated ISO
RAWMagick fully supports all cameras based on non-rotated RGB Bayer sensors.
Colour is before applying any profiles - it is out of the box colour. Applying colour profiles would distort histogram.
OK, OK, Neil. Much as I hate to admit it, you are correct.... :roll: The Canon has better detail in several areas, including the grass. I still think that the colors in the X lose more vivid. 8)
Dear Frank, it is not full demosaicing
What a mystery! I've got some questions about this...
First I was looking at the photos. There is certainly more shadow and highlight detail in the Nikon's shot, but the blacks are also lighter and whites less bright. Hmm, is the exposure the same?
So then I looked at the histogram and the applied (?) curve. The black point is much closer to the tail of the curve in the Canon shot, and the slope in the shadows is greater. But the biggest difference is the location of the white point! What is the logic behind these two radically different treatments of the high end of the curve?
Can you show us the curving on a shot taken with the Canon that utilizes the high end, and what would the same exposure do with the Nikon? Blow out or hold?
I'm not a RML user cause it doesn't run on a Mac... yet) but, using Photoshop, I rarely cut back the high end that much except when trying for an effect.
I agree w/ everything else you said (and was pretty much what I was thinking, but from a different perspective ), but shouldn't the top/right endpoints of the user curves be more similar than they are if DR is indeed very close? What would you call "very close"?
Perhaps, the diff is maybe just 1 stop out of 7-8?
Good point. I was starting to wonder about that also. Makes perfect sense.
Good to know even though I've been a bad boy and haven't gotten around to actually using it yet -- been sidetracked too much.
Look more carefully in the shadows and lower midtones, eg. in and around the doorway/door and adjacent areas. You'll probably need to flip between the two images to see clearly. It's arguable that the diff is actually quite a bit more noticeable than all the (low) "noise" (or even higher DR) diff that we hear about from the Canon folks.
As Iliah suggests, you can get the colors to be whatever you want (more or less), but you won't be able to recover the shadow details and will also have to deal w/ the "curve bending", which can lead to more tonal distortions/artifacts, etc.
The higher contrast of the Canon image will appear to have more "pop" and appeal at first (and may possibly be what you want for this image), but you could've gotten that from the Nikon image just as easily though. But the details and fine tones are gone forever -- and those things might reveal themselves more over time as you revisit the image (assuming you like it enough to do that of course ).
Hmmm... As I think a little more, I wonder what would happen if the Fuji S3 folks got a hold of this kind of look into the DR matter. Well, I guess most of them would probably just keep it to themselves and appreciate the fact that they have greater, truer DR and/or finer tones at their disposal as most of them are not the loud obnoxious sort to go duke it out w/ the Canon crowd. :wink:
Here are untouched by curve...
Dynamic range and coding it into 12 digital bits are quite different things. Say, I have noisy 3 least significant bits, and i know that. So, I will try to convert using upper 9 bits. But if I have analog noise equivalent to 2 bits, I will convert using 10 bits for the useful signal. Still I'm converting same dynamic range - but to different number of useful bits. Bits are actually steps in a ladder - if the distance between the steps is huge, I can get to 3 feet in just three steps; but I can make the ladder same 3 feet tall with 6 steps. Height of the ladder is dynamic range, number of steps is bits. Less steps - more posterization, noise, banding. Look here (photo and raw conversion by Miljenko Devcic):
Canon EOS 350D DIGITAL
So. Illah, if one exposes properly and then develops the image appropriately one should:
1. find Nikon files from D2X to be as capable to render as much details as the 1DsMkII
2. Nikon files to yield greater DR.
I think one problem is many are not developing their Nikon NEFs to their full potential (and not knowing that to be the case).
Based on the findings here, the Nikon files won't yield greater DR, but they will yield finer tonal steps across the DR, particularly in the shadows and midtones (at least the lower midtones). That means details in those regions are better preserved than the Canon and not as easily blocked up.
Please consider disabling your ad blocker for our website.
We rely on ad revenue to pay for image hosting and to keep the site speedy.
Or subscribe for $5 per year to remove all ads and support our efforts.