Thoughts on Medium Format Cameras

[Ray]

Hi,

I don't think that either the P65+ or the Canon had blown out highlights. Ray may be correct regarding the one stop difference in exposure. The histograms in Lightroom indicate this but Adobed has some bias that may falsify the comparison.

Best regards
Erik

The image comparison I'm referring to copied is below. The sky in the P65+ shot is clearly blown. Even the highlights on the metal sheeting on the far right are blown in the blue channel, at 249,253,255. At a similar location on the 1Ds3 shot the reading is 143,160,188. Big difference!

How anyone can present such comparisons and expect to be taken seriously, beats me.

[tho_mas]

The sky in the P65+ shot is clearly blown.

of course it is... the image was pushed by at least 5 stops in post. This was just to show how much more "content" you can recover in the shadows from the P65+.

[Ray]

of course it is... the image was pushed by at least 5 stops in post. This was just to show how much more "content" you can recover in the shadows from the P65+.

It is not necessary to blow the highlights in order to reveal shadow content. If anyone wishes to demonstrate that the P65+ has greater than a one stop DR advantage over the 1Ds3, the first requirement is to demonstrate that the images being compared have been equally exposed with regard to an ETTR.

To do this it is necessary to bracket exposures with both cameras at their minimum shutter speed or EV intervals so that one can precisely match pairs of images that have equal highlight detail after conversion.


DXO do not give DR comparisons at ISO 50 for these two cameras. What their results state is that the DR of the P65+ at the manufacturer-specified ISO of 100 is one stop greater than the DR of the 1Ds3 at its manufacturer-specified ISO of 100. They also claim that the DR of the P65+ at ISO 200 (actually ISO 89) is about the same as the DR of the 1Ds3 at ISO 100 (actually ISO 73).

It should not be difficult to confirm or refute such claims by DXOMark.

One also needs to be mindful of T-Stop considerations as well as true ISO sensitivity. One can't assume that all lenses at F11 produce equal exposures at the same shutter speed and the same ISO.

I should add that I am not disputing that the P65+ has a noticeably greater DR than the 1Ds3. Viewing Erik's screen shots showing the histograms of the bridge in Lightroom, it is clear that neither image is overexposed, but also that neither image is an ETTR. However, the 1Ds3 image clearly has a darker sky at zero settings.

The point I'm making is that these images do not suggest to me that the DXOMark results are wrong in practice. If the P65+ at ISO 100 has one stop better shadow detail than the 1Ds3 at ISO 100, as claimed by DXO, then it is reasonable to assume it has 2 stops better DR at ISO 50, than the 1Ds3 at ISO 100.

Add in the factor that the 1Ds3 shot looks slightly less exposed than the P65+ shot in this comparison, then one would expect the P65+ shot to have something greater than a 2 stop DR advantage in the shadows, which appears to be the case, thus tending to confirm the accuracy and relevance of the DXOMark test results.

[Ray]

Having had time to reflect on this matter, since my previous post was made late at night in Australia, it may well be the case that ISO 50 for both cameras produces exactly the same as result as ISO 100. Perhaps the only benefit of ISO 50 is the slower shutter speed that can be used.

If this is the case, that fact should be mentioned and some reason given for comparing the 1Ds3 at ISO 100 with the P65+ at ISO 50.

For those of us who are primarily interested in unbiased comparisons (as far as it's possible to be unbiased) we also need to know how DPP, ACR or Lightroom compare with C1 when converting Canon files. It is my experience that certain RAW converters, with zero or default settings, can provide more detailed results but at at the expense of noticeably greater noise.

It's possible that such factors can work together to give the impression that P65+ images have a greater-than-one-stop DR advantage over the 1Ds3. For example, perhaps it's the case that C1 applies an automatic noise reduction to Phase files which are inherently sharper due their lack of an AA filter, but don't apply the same noise reduction to Canon files.

If one underexposes the Canon shot slightly, uses a converter that brings out the maximum detail in the Canon file at the expense of increased noise, uses ISO 100 with the Canon camera instead of ISO 50, it may be quite possible to give the impression that the DXOMark results are misleading and that the DR difference between these two cameras is greater than DXO test results imply.

[tho_mas]

In RAW Developer the Canon file doesn't look any better.
I think one thing is pretty obvious: even if you deduct the 1 stop advantage for the P65+ due to the real ISO (you are right here) you can beat the file around much more than the Canon file... there is simply more (usable) information. 3-4 stops maybe... I don't know, but clearly much more than 1 stop.

[Ray]

In RAW Developer the Canon file doesn't look any better.
I think one thing is pretty obvious: even if you deduct the 1 stop advantage for the P65+ due to the real ISO (you are right here) you can beat the file around much more than the Canon file... there is simply more (usable) information. 3-4 stops maybe... I don't know, but clearly much more than 1 stop.

It seems that none of us knows, or is keeping it a secret. The comparison under discussion is deeply flawed because of an apparent lower exposure of the 1Ds3 image and a difference in ISO sensitivities.

Of course a larger file consisting of a significantly greater number of pixels of similar DR characteristics will lend itself to more intense processing before posterization and other image degradation artifacts become apparent.

Simply downsizing the P65+ file to the same size as the 1Ds3 file increases the DR advantage from 1/4th of a stop to one full stop.

The 16 bit A/D converter in the P65 will provide some advantage over the 14 bit A/D converter of the 1Ds3, as regards post processing, just as it's better to convert a RAW file into 16 bit as opposed to 8 bit color for processing purposes. It's even advisable to convert an 8 bit TIFF original into 16 bit if it is intended to 'beat the file around', as you put it.

As regards DR, I imagine the post processing advantages of an original 16 bit file with almost 3x the pixel count will contribute a marginal increase to that 1 stop advantage shown in the DXOMark test results.

Perhaps a more significant factor, which DXO does not address, is the additional noise in the 1Ds3 image which might result from the greater degree of sharpening required for an image which has been subject to the effects of an AA filter.

On the other hand, the skillful operator will protect the deepest shadows when sharpening, as one is able to when using Photoshop's 'Smart Sharpen'.

I would be prepared to accept that certain P65+ images which require a lot of post processing might show a 1 & 1/2 stop DR advantage over the 1Ds3, at normalised print sizes.

But claims of a 3-4 stop advantage are simply not credible.

[hjulenissen]

The 16 bit A/D converter in the P65 will provide some advantage over the 14 bit A/D converter of the 1Ds3, as regards post processing, just as it's better to convert a RAW file into 16 bit as opposed to 8 bit color for processing purposes. It's even advisable to convert an 8 bit TIFF original into 16 bit if it is intended to 'beat the file around', as you put it.
But claims of a 3-4 stop advantage are simply not credible.

Do you know this for a fact? I have seen some claims that the 15th and 16th bit contain only noise. If it does, then it should not give any benefit compared to converting a 14-bit file to 16 bits prior to editing.

In any case, the total quality is what is important, and numbers from a sales-brochure should not be as important as relevant as looking and measuring what is in the file.

-h

[Ray]

Do you know this for a fact? I have seen some claims that the 15th and 16th bit contain only noise. If it does, then it should not give any benefit compared to converting a 14-bit file to 16 bits prior to editing.

In any case, the total quality is what is important, and numbers from a sales-brochure should not be as important as relevant as looking and measuring what is in the file.

-h

No, I don't know that for a fact. It's difficult even to show the benefits in an image with a 14 bit A/D conversion from certain Nikon DSLRs, compared with the same scene shot in 12 bit mode using the same lens and camera.

Giving Phase the benefit of the doubt, I would expect only a very marginal IQ benefit from such increased bit depth at the A/D conversion stage. I understand also that Photoshop's 16 bit conversions from RAW files are effectively 15 bit, but I'm not going to lose any sleep over it.

As I mentioned, the greatest DR benefit from the larger files of the P65 are already shown on the DXOMark site, ie, an increase from 1/4th of a stop at the pixel level to a full stop when the P65 image is downsampled to the same size as the 1Ds3 image.

An increase in noise due to the extra degree of sharpening required of the 1Ds3 image, may have the effect of increasing that 1-stop DR advantage slightly.

It is clear in general that MFDBs have a slight resolution advantage in parts of the image that are not affected by aliasing, as a result of their lack of an AA filter.

For this reason, even when DoF considerations are not an issue, it would be considered sound practice to adjust the F stop as well as the focal length when comparing different format cameras. Everyone should know that a lens at F7.1 or F8 is sharper than the same lens at F11.

This current comparison from Phase has used F11 with both the P65+ and 1Ds3. Why?

This is tantamount to making a statement in bold letters, "This comparison is biased".

Comparing the 1Ds3 with the P45 it would be appropriate to use F8 with the 1Ds3 for the same FoV as the P45 shot at F11, using appropriately different focal lengths of lenses.

However, the P65 is a larger format than the P45, so in circumstances where F11 is used with the P65, one should use F7.1 with the 1Ds3, if one wishes to even attempt to be objective. Such an image at F7.1 would not need as much sharpening in order to match the resolution of the downsampled P65+ image.

If there are any P65+ owners out there who also have access to a 1Ds3, and who are serious about resolving this issue of the relevance of DXOMark test results, I can recommend the following procedure.

1. Select two, good quality, prime lenses of the same focal length, one that fits the 1Ds3 and the other that fits the MF camera used with the P65+.

2. Determine the T-stop of both lenses in order to avoid exposure confusion, and/or be prepared to bracket exposures in order to correctly match equally exposed frames from each camera.

3. Set the P65+ at ISO 200 (ISO 89 according to DXOMark). Set the 1Ds3 at ISO 100 (ISO 73 according to DXOMark). This is the closest match of ISO sensitivity you can get.

4. Shoot the same high-contrast scene under constant lighting conditions, from the same position, using the same F stop with both cameras.

5. Do not be confused by any significant differences in exposures that might not appear to match the slight differences in ISO sensitivity. The P65 image will have a wider FoV that might include brighter parts of the scene that are not included in the FoV of the 1Ds3.

This is why it is helpful to know the T-stop of both lenses as a starting point for bracketing.

6. Crop all the bracketed P65+ images to the same FoV as the 1Ds3 shots.

7. After cropping (and not before) match a pair of images, one from each camera, that mostly closely have the same exposure and the same highlight detail..

8. Adjust WB and levels so that both images appear similar in tonality and hue.

9. Lighten the deepest shadows equally in both images, and compare.

Now the reason for this rigmarole is in order to compare images at the pixel level, equivalent to the 'screen' mode on the DXOMark website. What DXO tell us is that the 1Ds3 at ISO 73 has approximately 2/3rds of a stop better DR than the P65+ at ISO 89, at the pixel level or screen mode.

2/3rds of a stop better DR should be noticeable if there are deep shadows in the scene that has been photographed.

If the cropped P65+ image, in these circumstances, shows significantly better detail and less noise in the shadows, then this would be good evidence that the DXOMark results have questionable relevance as regards MFDB performance.

If shadow detail appears about the same, one might have to consider if the shadows are really deep enough to reveal the DR differences. I've noticed a lot of confusion recently about the significance of an input signal of 1% on the log scale. Some people seem to think such a level is representative of deep shadows, but DXO graphs suggest otherwise.

One might also consider that manufacturing and quality control differences might be the cause of any slight variance with the DXO tests. Whilst a 2/3rds of a stop difference in DR should be noticeable, it's not particularly significant.

I await with bated breath for someone to conduct such a comparison. 

[Jim2]

If one does not mind stitching a large number of images, the D7000 might be preferable to the D3x, since the former has a higher pixel density (4.73 vs 5.9 microns) and one can use the central portion of the projected image where optical performance is higher.

If I read this correctly, you're saying that sensors with smaller photo cell size is better? A cropped image from D3x will be worse than the same image from D7000 assuming the cropped d3x image is of the same scale as the image captured by d7000?

[Ray]

If I read this correctly, you're saying that sensors with smaller photo cell size is better? A cropped image from D3x will be worse than the same image from D7000 assuming the cropped d3x image is of the same scale as the image captured by d7000?

The essential points here are that the D7000 pixel, although smaller than the D3X pixel, has about 1/2 a stop greater dynamic range but is approximately equal with regard to the other parameters that DXOMark test, such as tonal range, SNR at 18% grey, and color sensitivity. A half stop DR advantage is considered to be noticeable and worth having, but not particularly significant.

A row of 3 stitched D7000 images, with camera vertical, cropped to the same FoV and file size as a single shot with the D3X, camera horizontal, should have marginally better DR than the single D3X image, and noticeably better resolution in the corners and at the short edges, if a full-frame lens has been used.

If a single D3X image is cropped to the same FoV as a single D7000 image, having used the same lens of the same focal length with each camera, one would expect the cropped D3X image to be technically inferior.