1Ds2 versus 5D

[bjanes]

Here is a 100% pixel crop, with the 5D image (left) slightly interpolated (PS2&Bicubic sharper) to match the 1DsII:
Cheers,
Petter
[a href=\"index.php?act=findpost&pid=70345\"][{POST_SNAPBACK}][/a]

Petter,

A nice test. I preferred the 100% view of the 5D, since the contrast of the letters appeared to be better, but the difference is quite small. Your choice of bicubic sharper interpolation is interesting. The official Adobe recommendation for upresing is bicubic smoother, but for poster sized prints Scott Kelby in his Professional Photoshop for Photographers recommends bicubic sharper. Confusing    

Bill

[jani]

Sounds like there might be some misfocussing going on. It's surprising how critical focussing can be with a long lens.

Not really, considering how critical it is at f/1.4 with a 50mm lens.  

When comparing camera bodies with the lens fixed to the tripod, I always remove the body from the lens, leaving the focussing the same.

Yes, the same here.

However, comparing 400mm with 560mm might require refocussing. I'm not sure about this, but that's what I've been doing, and occasionally I find that focussing is then perceptibly different.

We took several shots with and without refocusing, so it appeared to be very consistent, although the differences were rather small.

[Jack Flesher]

Since I own both bodies I decided to do a (very) quick test. I don't know if you feel it helps the discussion, but here it is...

SNIP

Cheers,
Petter
[a href=\"index.php?act=findpost&pid=70345\"][{POST_SNAPBACK}][/a]

Nice example Petter.  However your image (like the crop of the dollar bill used at the beginning of this thread) is one of significant contrast, medium-frequency detail.  I suspect even the 20D would hold its own compared to the 1Ds2 in this example and have all of us claiming that we really don't need any more detail than the 20D offers

I suggest you repeat the experiment with a subject that has more subtle tonal transitions in it (lower contrast) and even higher-frequency detail.  Maybe try a shaded bush with smallish (2 - 3 cm) leaves at about 20 meters distance using a good prime lens at f8.  

Cheers,

[PetterStahre]

In response to various posters...

Dwdallam, Ray and Jack... thanks a lot for the feedback. Here comes a new shot which I hope is what you're looking for.

Skid00skid00... the color error from the 5D (and 1DsII) hasn't to do with lens aberation it seems. But if I activate just a little anti-noise (as I would normally) it all disappears.

Bill... I normally prefer Bicubic smoother for large interpolations. Bicubic sharper tends to show the pixels quite fast, but I wanted to interpolate the 5D just a little (to match 1DsII) and that meant (to my taste) I could use the Bicubic sharper method (which is more visually appealing if not used to the extreme). (By the way, Scott Kelbys book is full of errors!)

I went out to shoot a city scene here in Stockholm this evening. I used the same cameras but with my 85/1.2 lens. One of you suggested using a small aperture but that lens is at it's sharpest at f4. (Read more at  http://www.photozone.de/8Reviews/lenses/ca...85_12/index.htm )
And since it was a little bit windy I wanted a fast shutter speed to reduce risk of camera shake or leaves being blurred. (I also shot using f5.6 + f9 but couldn't see any difference after spending half an hour pixel peeping so for this post it's the f4-images I use.)

I used a tripod, mirror lockup, cable release and auto focus set on one of the buildings some 150 meters away (meaning I had DOF approximately from 40-50 meters to infinity with f4).

I used auto white balance and didn't touch it in RAW Developer (so the small differences in color is mostly due to this I believe). ISO 100 and Av-program (I got 1/800 for 5D and 1/640 for 1DsII, both using evaluative metering).

This is the scene I shot:


And to my eyes I wouldn't say the 5D has better dynamic range... rather the opposite! But maybe 5D can be measured to have better DR!? For me personally I would say they are equal!

Moiré is present in various parts of both images. It just occurs on different areas due to the native resolution differences.

// Petter

[Jack Flesher]

Nice test Petter:

I think the leaves in #2 and #3 (the closest to the center of the lens, and likely the sharpest area for the image) begin to tell the story -- especially in the high-frequency detail like the leaves, brickwork and fence grating.  I believe you will see even more difference at f5.6 or 8 with that particular lens.  

However, as we all are aware, applying NR kills off high-frequency (micro) detail in a hurry!   If you have the time, I would like to see each of these same crops side-by-side again without any NR run during conversion...  I supect we will see a larger difference in detail if you do so -- along with the color errors and a more accurate comparison of noise in the 6 crop.

To DR I saw about 1/2 stop advantage to the 1Ds2 when I did my comparison and never agreed with the 5D has more DR crowd -- your highlight example shows maybe 1/3 stop and it's hard for me to tell what's going on in the shadows as they look very close -- maybe also due to the NR you ran. Though slightly more fine detail is visible in the 1Ds2 image (bricks).

Are these significant differences?  Probably not for many, but certainly for some.  I would run them both through a second competitive converter like RawShooter  -- and certainly compare them with NR off -- before I made any concrete conclusions.  

Now to the color errors on the 5D -- ringing any bells of recognition with you Ray?        As you recall, I commented on this very subject (5D color-cast problem) in the lengthy RG thread about 6 months ago...  Fact is, the 5D shows a color cast across the sensor in certain situations.  I did not keep mine long enough to determine exactly what conditions cause it, but for sure it is real.  FWIW I have never seen it on my 1Ds2.

Petter, thanks again for your efforts on this!

[PetterStahre]

...snip...
I believe you will see even more difference at f5.6 or 8 with that particular lens. 

However, as we all are aware, applying NR kills off high-frequency (micro) detail in a hurry!
...snip...

I would run them both through a second competitive converter like RawShooter  -- and certainly compare them with NR off -- before I made any concrete conclusions. 

Thanks for the feedback.

About the aperture. No, as I wrote  I also spent some time evaluating f5.6 and f9 but they gave me no more details (even if there are very, very subtle differences between the images).

Here is an example (1DsII f4 vs f9, no NR and no USM):


About NR... same thing here. Here's an example (no NR vs NR, but with USM):


And here's a comparision between RAW Developer and ACR (no NR but USM):


Cheers,
Petter

PS. I also ran tests both before my previous post and now, using no NR for the DR-test... no difference of interest to report.

[Jack Flesher]

SNIP

PS. I also ran tests both before my previous post and now, using no NR for the DR-test... no difference of interest to report.
[a href=\"index.php?act=findpost&pid=70550\"][{POST_SNAPBACK}][/a]

But it would be interesting to see the #3 (center) crop form the 5D and 1Ds2 side by side WITHOUT NR on...  

,

[skid00skid00]

Thanks for the clarification, and additional test shots, Petter.

[Ray]

Now to the color errors on the 5D -- ringing any bells of recognition with you Ray?        As you recall, I commented on this very subject (5D color-cast problem) in the lengthy RG thread about 6 months ago...  Fact is, the 5D shows a color cast across the sensor in certain situations.  I did not keep mine long enough to determine exactly what conditions cause it, but for sure it is real.  FWIW I have never seen it on my 1Ds2.

[a href=\"index.php?act=findpost&pid=70527\"][{POST_SNAPBACK}][/a]

I don't see any color cast issues with these test shots from Petter, Jack. The only differences in color I see in these shots is in the shadows crop of area 6. There's more magenta in the 1Ds2 shot (on the right), which I would say should not be there. However, I recognise that the 1Ds2 shot seems to have effectively slightly less exposure, as indicated by the greater detail in the bow of the boat with a -3 EC adjustment. Because the highlights are better, in the 1Ds2 shot in the crop of the boat, but the shadows better in the 5D image, in area 6, it's difficult to say which camera has the better dynamic range.

It is clear, however, in the area 2 shots that the 1Ds2 has the resolution edge. It's not just the leaves that look better in the 1Ds2 shot, but acrros the middle of the frame, above the foliage, there's some design pattern that looks a bit like Chinese script. This is clearly better defined in the 1Ds2 shot.

So for me, the matter is resolved. The 1Ds2 clearly does have a slight resolution edge, as indeed one would expect it to have with 30% more pixels.

The message for me, here, is that 30% more pixels is not substantial. If the 1Ds3 has another 30-40% more pixels, making it around 22mp, I suppose the danger for sales will be new owners of the MK lll will compare resolution with the MK ll and find the same sorts of differences that we are now seeing between the MK ll and the 5D.

However, then we shall be in a position where there will be no contest between the new 1Ds3 and the old 5D. Two small differences make a big difference   .

[Jack Flesher]

I don't see any color cast issues with these test shots from Petter, Jack.
SNIP

[a href=\"index.php?act=findpost&pid=70622\"][{POST_SNAPBACK}][/a]

That's because as he said, he ran NR to remove them...  As I said above *twice*, I would like to see the comparison crops so we can all see what he mentioned in his first post, but has not shown.

,

[Jack Flesher]

And here's a comparision between RAW Developer and ACR (no NR but USM):


[a href=\"index.php?act=findpost&pid=70550\"][{POST_SNAPBACK}][/a]

FWIW, I don't feel ACR is a competitive raw converter for extracting detail from 1Ds2 files  

[Ray]

That's because as he said, he ran NR to remove them...  As I said above *twice*, I would like to see the comparison crops so we can all see what he mentioned in his first post, but has not shown.

,
[a href=\"index.php?act=findpost&pid=70628\"][{POST_SNAPBACK}][/a]

A problem that can be solved or removed is no longer a problem.

[Jack Flesher]

A problem that can be solved or removed is no longer a problem.
[a href=\"index.php?act=findpost&pid=70636\"][{POST_SNAPBACK}][/a]

It is if it creates other problems -- and NR kills detail, plain and simple.  

But I certainly respect that you may not care about giving up detail to eliminate a color problem.

[Ray]

It is if it creates other problems -- and NR kills detail, plain and simple. 

[a href=\"index.php?act=findpost&pid=70640\"][{POST_SNAPBACK}][/a]

True! But noise reduction is not necessarily the best method of fixing a color shift, is it? Surely there's a whole range of options available in PS for changing color without affecting resolution. Let's not ceate problems where none exist   .

[Jack Flesher]

True! But noise reduction is not necessarily the best method of fixing a color shift, is it? Surely there's a whole range of options available in PS for changing color without affecting resolution. Let's not ceate problems where none exist   .
[a href=\"index.php?act=findpost&pid=70650\"][{POST_SNAPBACK}][/a]

Ray, clearly you do not understand the nature of the problem I (and others) have seen with their 5D sensors.

It is not a set color shift but a color shift across the sensor, usually from cyan-green to magenta, originating in one corner with the cyan cast and ending in the opposite corner with a magenta cast.  This particualr issue is NOT easily corrected in post-processing.

[Ray]

It is not a set color shift but a color shift across the sensor, usually from cyan-green to magenta, originating in one corner with the cyan cast and ending in the opposite corner with a magenta cast.  This particualr issue is NOT easily corrected in post-processing.
[a href=\"index.php?act=findpost&pid=70666\"][{POST_SNAPBACK}][/a]

Haven't noticed it, Jack. But now I'm going to start looking for it. If I find it, I'll let you know   .

[Jack Flesher]

Haven't noticed it, Jack. But now I'm going to start looking for it. If I find it, I'll let you know   .
[a href=\"index.php?act=findpost&pid=70669\"][{POST_SNAPBACK}][/a]

Frankly, it's a non-issue for me   (Plus when and if you do find it, I'm sure it won't matter to you!)

Cheers,

[PetterStahre]

That's because as he said, he ran NR to remove them...  As I said above *twice*, I would like to see the comparison crops so we can all see what he mentioned in his first post, but has not shown.

You don't trust strangers, do you?  

Seriously, I understand your request since with these kind of tests it's easy to make small errors. And there is also a certain amount of subjectivity involved when viewing the results. So here it is...

First, to underline why I used NR, here is the 1DsII without NR (left) and with NR. No USM in either image:
 

Hope this helps.

Cheers,
Petter

[Jack Flesher]

You don't trust strangers, do you?  

[a href=\"index.php?act=findpost&pid=70703\"][{POST_SNAPBACK}][/a]

Has nothing to do with trust -- I think different sets of eyes see things differently, that's all.  

For example, even in the new jpeg web crops you posted immediately above without NR but with USM, to my eyes the 1Ds2/5D comps show a greater difference in detail than those you posted ealrlier with NR applied;  I see essentially no difference to the 5D file with NR applied compared to without, but the 1Ds2 file without NR looks to have a more detail than the NR version.  

My .02 to be sure and  I'm equally sure others will disagree  

Thanks again for all of your work on this -- I think these are fine examples.

Oh, and I do agree there is no serious color issue visible with either camera

[PetterStahre]

Has nothing to do with trust -- I think different sets of eyes see things differently, that's all. 

I agree.

However I saw I've made a mistake... about the first image comparision in my last post (1DsII "NR vs no NR")... the image without NR is the one to the left. In my original posting i mentioned the opposite but has now edited it.