I want to pre-empt an inevitable controversy.

BE WARNED, there is certainly going to be a lot of gnashing of teeth and wailing about how noisy and unsharp the 5DmkII is after it ships, and people start comparing 100% crops with their old 5D's. Ignore it. This will only happen because people will fail to uprez their 5D files to 21 megapixel before comparing 100% crops.

Even reviewers like DPReview and Luminous Landscape routinely make this methodological error when comparing cameras.

There is a simple way to see what your new files are really like: uprez the 5D file to 21 megapixels, and then compare 100% crops. That way the images are magnified to the same size, and the comparison is fair. Contrary to popular belief, uprezzing does not turn a unicorn into a pantomime dromedary.

Lenses like the 24-105 are unlikely to take full advantage of the 21 megapixel sensor. Depending on the aperture, some significant part of the frame will be resolved at less than 21 megapixel resolution. How much that matters to you depends on what kind of photography you want to do.

I have been using the 1Ds3 since December, and although I no longer own a 24-105, I use a Zeiss 24-85 N which is quite similar, and with quite a few other lenses also I have noticed that they limit sharpness, and require more sharpening in post. However, some lenses really milk the sensor and require no abnormal sharpening, so it's definitely not the AA filter.

Primes generally give good results. Expect to use a 17-40 at f8 or f11, with DPP lens aberration correction, to get optimal results. Even then, you will see some loss of sharpness towards the corners. Here's an example:

great gif! the woman in white seems to have a gerbil in her purse.

I am Daan B, and I approve this message

A good point, and well made.

Question: Why not just view the two images at the appropriate levels of scaling so that objects are the same size in both? For test purposes it would be easy to include a ruler or something to set the scale and it would avoid any processing artefacts introduced by the upscaling.

15Bit wrote:
A good point, and well made.

Question: Why not just view the two images at the appropriate levels of scaling so that objects are the same in both? For test purposes it would be easy to include a ruler or something to set the scale and it would avoid any processing artefacts introduced by the upscaling.


That's a good question. The problem seems to be that people are not happy with the less than optimal way that programmes display images at non-native resolutions. Sometimes you can see weird patterns appear in photoshop when a file is displayed at, say, 91%. Or is it the operating system's in-built graphics sub-system? Or the graphics card. I don't know, but the consensus seems to be that it is not fair to set the 5D file's magnification to 130% in photoshop, as the 21 megapixel file will have an advantage at 100%. Uprezzing using bicubic is a very good conservative approximation which seems barely to damage detail at all. Before the pantomime dromedaries canter in, there are more ways of resizing in order to make a fair comparison, as follows.
- Uprez the lower megapixel file to the higher file's dimensions using bicubic. This is the quickest and most sensible in my view.
- Downrez the higher megapixel file to the dimensions of the lower. This penalises the higher rez file by discarding some arbitrary amount of detail.
- Uprez both files bicubically to some higher rez common denominator. Seems fair, but I contend, pointless.
- Preserve the lower rez file perfectly by doubling exactly the baseline (quadrupling the number of pixels) using nearest neighbour, and uprezzing the higher rez file to match using bicubic. This answers the pantomime dromedaries, since the lower rez file doesn't change at all, and it remains mathematically, logically identical to the original.

I have tried all these methods, and have never seen a difference in detail comparison except in the second, where detail is obviously discarded. I recommend the first, for sanity, although it may not be so fair when file sizes are very close. That's not the case with 5D and 5DmkII.

The 100% crop concept is garbage in so many ways. If you do want to peep, you need to up-rez or down-rez. I take ISO3200 images on my 30D with a 24-70 run through Lightroom's standard noise controls. They don't look that much at 100% but for modest prints you'd never know.
One other mistake is somone with a 10 or 8MP camera posting, say, a face from a full-length portrait and saying it's not sharp while an eye from a tightly cropped portrait is deemed 'sharp'.
Psychologically, we want to see a whole face sharp but are willing to see an eye less sharp as we see that as an 'extreme enlargement'.
I can't believe we're still having this argument. I remember a comparison of the D100 and D200 where they uprezzed from 6MP to 10MP to compare. The D200's image had a little less 'haze' at 100% but you had to use imagination to see any extra detail.

I take the shot, make sure it's in focus, and sharpen appropriate for where the image will be output - web, inkjet, lab. I couldn't give a monkey's beyond that.

ShutterLover wrote:
The 100% crop concept is garbage in so many ways.

Well, it's useful for choosing between two shots taken in succession on the same camera, especially if there is a focus, shake or subject movement difference, and if you resize files to the same dimensions it's useful for comparing output from two different cameras. If you're going to use a camera as a tool for a few years, those marginal differences can be important. But I agree with you that it is a higher level of scrutiny than most small prints will ever get, and per pixel noise is seldom resolved by printers even in bigger prints.

brainiac wrote:21 megapixels may be nearly double the number of pixels you are used to, but it is nowhere near the 52 megapixels that would double the 5D's resolution.

Quadruple, not double.

I vote this most useful topic of the week.

(equal alogside the "HIDE ME" button)

alexanderino wrote:
brainiac wrote:21 megapixels may be nearly double the number of pixels you are used to, but it is nowhere near the 52 megapixels that would double the 5D's resolution.

Quadruple, not double.

Which 5.25 megapixel camera are you using? Most 5D users are used to 13 megapixels, so the mkII will be about double the number of pixels. In order to double resolution, you would need to quadruple the number of pixels to 52 million. Read what I said more carefully.

This is certainly the case when reviewing "crops" and digital files.
In print, this is not the case, and it is good to see you make the distinction.
In print, 12mp on a 16X20 VS 21mp in a 16X20 makes big difference; which becons the question whether or not the print may be loosing its importance as a final medium...I am not sure.

Good post though....however, I think we are not going to be immune from these upcomming comparisson posts he he he.

dasrocket wrote:
I think we are not going to be immune from these upcomming comparisson posts he he he.

I know it - but we must fight the fight. Remember to boycott DPReview and Luminous Landscape since they compare crops at different magnifications and make judgements on per pixel noise.

Better still, email them explaining why you disdain their incompetence at comparing crops at different magnification, and ask them to apologise to those they have misled, to do fair comparisons next time, and correct the methodological error that is extant all over their sites.

brainiac, you are a heretic. how dare you introduce such doubt to the sheeple? whatever will they buy?!

Contrary to what Dasrocket opines, I suspect most will notice very little difference, if any, between a 16 x 20 from either camera of a similarly exposed scene. I say this because it is true when comparing my 30D to my 1DSMK2 and that is about double the megapixels.

Some go to great lengths to convince themselves (and try to convince others) that throwing huge sums of money at a product somehow translates to huge differences in quality on the print end. It doesn't. While there are reasons to upgrade, a substantial difference in the final print is not high on the list for a lot of our "everyday" shooting.

John Power wrote:
Contrary to what Dasrocket opines, I suspect most will notice very little difference, if any, between a 16 x 20 from either camera of a similarly exposed scene. I say this because it is true when comparing my 30D to my 1DSMK2 and that is about double the megapixels

As a user of 5D and 1Ds3, I suspect you are wrong, but they will need to shoot carefully with the right lens. There is a tangible improvement in detail and crispness. That alone is not worth upgrading for, but ISO 25600, 14 bit, Liveview, the monitor, 4 fps, sensor cleaning, 3 C-modes, UDMA, and 1080p video all contribute to making it a worthwhile upgrade, and the extra detail is just a bonus.

brainiac wrote:
John Power wrote:
Contrary to what Dasrocket opines, I suspect most will notice very little difference, if any, between a 16 x 20 from either camera of a similarly exposed scene. I say this because it is true when comparing my 30D to my 1DSMK2 and that is about double the megapixels

As a user of 5D and 1Ds3, I suspect you are wrong, but they will need to shoot carefully with the right lens.

We have been down this road before (along with Daan) and our respective positions are well documented so I will not belabor the point any longer.

I thought you sold that 1DSMK3

It's still for sale, although I'm considering hanging on to it until the 5DmkII price gets rebated. The 1 series AF and speed is, ahem... enjoyable, and I don't get the impression so far that the 5DmkII is going to improve much on the 1Ds3's ISO 12800 performance:

John Power wrote:
Contrary to what Dasrocket opines, I suspect most will notice very little difference, if any, between a 16 x 20 from either camera of a similarly exposed scene. I say this because it is true when comparing my 30D to my 1DSMK2 and that is about double the megapixels.

Whether you see the extra detail in print is also highly dependent on your printing process, printer, medium, and print size.

I've long admired this explanation, Brainiac. I have read it a few times, and it only makes plain sense. Like a cattle prod for pixel peepers. Nicely stated, as usual

I find it pointless to compare a 12.8 MP camera to a 21.1 MP camera...
this is like comparing a 100 hp car to a 200 hp car and expect the 100 hp car to do the same performance as the 200...apples and oranges

digitalbug30d wrote:
I find it pointless to compare a 12.8 MP camera to a 21.1 MP camera...
this is like comparing a 100 hp car to a 200 hp car and expect the 100 hp car to do the same performance as the 200...apples and oranges

It's actually a lot more like comparing a 100 hp car to a 130 hp car, in a 70 mph speed limit zone.

brainiac wrote:
John Power wrote:
Contrary to what Dasrocket opines, I suspect most will notice very little difference, if any, between a 16 x 20 from either camera of a similarly exposed scene. I say this because it is true when comparing my 30D to my 1DSMK2 and that is about double the megapixels.

Whether you see the extra detail in print is also highly dependent on your printing process, printer, medium, and print size.

Not to be argumentative, but what is the point of going for more megapixels, if there is little to no difference on print? What are we gaining besides bigger files to handle and store?