Nikon is NOT on crack - Initial D3x image quality is AMAZING!

[Ray]

I imagine the argument is getting too technical for many of us. We need specific examples of real-world images of the sort photographers make (but of stationary subjects with constant lighting) comparing for example downsized D3X images with full size D3 images, or downsized 50D images with full size 40D images. The processing steps, resampling method and sharpening amounts should also be specified.

If the 50D image has been downsampled using 'bicubic sharper', which I also use when downsampling, then one might expect the downsampled 50D image to have slightly greater accutance than the full-size 40D image. If this proves to be the case, then there is no argument to be made that the 50D image still appears to be slightly noisier, since the whole purpose of the exercise is to trade resolution for lower noise. If the 50D image is still marginally sharper, but also marginally noisier, the trade-off is not complete.

I find the argument that one doesn't buy a camera with a higher pixel count in order to resample the images, disingenuous. One buys a camera to take photos, and the choice of presentation size of such images will depend on many factors such as the size one's printer can handle, the resolution of the projector or display, the degree of objectionable noise apparent if a high ISO was used, the extent of cropping required, and so on. Those who used the first 3mp DSLR that Canon produced (the D30), seemed to have been very happy with the quality of A4 size prints, and even A3 size prints, which of course involved very substantial upsampling. The work of many professionals using 39mp MFDBs will often end up being downsampled to a size suitable for magazine publications.

The lower the camera's pixel count, the more frequently images will be upsampled. The higher the camera's pixel count, the more frequently images will be downsampled.

I happen to be in the situation of owning both a 40D and 50D. Would there be any reason whatsoever, in any circumstances, for me to pick up the 40D instead of the 50D when going out shooting, apart from the trivial reason of being able to fit more images on the card? Is it really necessary for me to go to the trouble of making comparisons at various ISOs and waste ink and paper printing out the results?

[ejmartin]

I imagine the argument is getting too technical for many of us. We need specific examples of real-world images of the sort photographers make (but of stationary subjects with constant lighting) comparing for example downsized D3X images with full size D3 images, or downsized 50D images with full size 40D images. The processing steps, resampling method and sharpening amounts should also be specified.

It's not a perfect example, there are focus and exposure differences, but for noise it's reasonable to use the Imaging-Resource raw files.  I did that with the 40D and 50D at ISO 1600.  Here they are at native resolution, converted by DPP with default settings for each camera, except for turning off NR:

http://theory.uchicago.edu/~ejm/pix/20d/po...40DhSLI1600.jpg
http://theory.uchicago.edu/~ejm/pix/20d/po...0DhSLI01600.jpg

and the 50D resampled to the pixel dimensions of the 40D using PS Bicubic

http://theory.uchicago.edu/~ejm/pix/20d/po...00-downsamp.jpg

I did the same exercise with the Imaging-Resource multitarget images, and had someone run them through Imatest.  The downsampled 50D image had slightly higher resolution, but nothing dramatic and perhaps within the error bars.  The downsampling mitigated the interpolation artifacts in the 50D image so that they were less objectionable than the 40D at its native resolution.  Lanczos is a more accurate resampling filter, and results in slightly less noise than PS Bicubic.

The resampling of the 50D is merely a sideshow IMO, to show using onscreen display the relative noise at equal output size.  The noise levels at comparable spatial scales are equal even prior to resampling, as one can see from the noise spectra.

[Ray]

It's not a perfect example, there are focus and exposure differences, but for noise it's reasonable to use the Imaging-Resource raw files.  I did that with the 40D and 50D at ISO 1600.  Here they are at native resolution, converted by DPP with default settings for each camera, except for turning off NR:

Emil,
Thanks for saving me the time and trouble of doing such comparisons myself. I went to some trouble comparing 40D and 50D resolution, photographing a banknote, but was quite sure that 50D noise would be no worse than 40D noise, when comparing equal size images, so I did not bother to shoot any high-ISO images for comparison. I generally only make tests when I have reason to doubt what is claimed.

Out of curiosity, I upsampled the 40D image to the 50D size, using PS bicubic, then sharpened the interpolated image with Focus Magic using a blur width of 1 pixel and an amount of 50%.

At 200% magnification on the monitor, the text on the beer bottle is clearer in the 50D image, despite the fact that the 40D shot appears to have been taken from a slightly closer distance to the scene (or the focal length was slightly greater). The slightly greater resolution of the 50D is consistent with my own tests, but the magnitude of the resolution difference is a bit disappointing and highlights the fact that improvements in lens design and manufacture are not keeping pace with improvements in sensor technology.

[attachment=10589:40D_upsa...50D_size.jpg]

What we need now is a similar comparison between the D3 and D3X.

[Theodore]

Back to the topic of initial impressions regarding the D3x image quality, Lloyd Chambers has been shooting with one and while his observations are on his paid-access site, he does post some general impressions on his blog:  http://www.diglloyd.com/diglloyd/blog.html  For those who haven't checked out Lloyd's site, there's a lot of information to be found there regarding Zeiss lenses.

[MatthewCromer]

Emil,
Thanks for saving me the time and trouble of doing such comparisons myself. I went to some trouble comparing 40D and 50D resolution, photographing a banknote, but was quite sure that 50D noise would be no worse than 40D noise, when comparing equal size images, so I did not bother to shoot any high-ISO images for comparison. I generally only make tests when I have reason to doubt what is claimed.

Out of curiosity, I upsampled the 40D image to the 50D size, using PS bicubic, then sharpened the interpolated image with Focus Magic using a blur width of 1 pixel and an amount of 50%.

At 200% magnification on the monitor, the text on the beer bottle is clearer in the 50D image, despite the fact that the 40D shot appears to have been taken from a slightly closer distance to the scene (or the focal length was slightly greater). The slightly greater resolution of the 50D is consistent with my own tests, but the magnitude of the resolution difference is a bit disappointing and highlights the fact that improvements in lens design and manufacture are not keeping pace with improvements in sensor technology.

[attachment=10589:40D_upsa...50D_size.jpg]

What we need now is a similar comparison between the D3 and D3X.

Hi Ray.

I think the image on the left is significantly more detailed than the one on the right.  I'm assuming the one on the left is the 50D?  A pretty good increase of resolution in my book.

[Ray]

Hi Ray.

I think the image on the left is significantly more detailed than the one on the right.  I'm assuming the one on the left is the 50D?  A pretty good increase of resolution in my book.

Yes, the one of the left is labelled 50D at the top. At 200% the difference is clear. If I were comparing lenses in the store before buying and I got this degree of difference, I would definitely buy the lens that produced the image on the left.

The problem is, if you were to make a print of the full image in order to view such differences from the same distance you view your monitor, the print would need to be about 6ftx9ft.

Nevertheless, if one is shooting wildlife using one's longest telephoto lens and the result needs extensive cropping, then the 50D will have a clear advantage over the 40D.

[JohnKoerner]

Yes, the one of the left is labelled 50D at the top. At 200% the difference is clear. If I were comparing lenses in the store before buying and I got this degree of difference, I would definitely buy the lens that produced the image on the left.
The problem is, if you were to make a print of the full image in order to view such differences from the same distance you view your monitor, the print would need to be about 6ftx9ft.
Nevertheless, if one is shooting wildlife using one's longest telephoto lens and the result needs extensive cropping, then the 50D will have a clear advantage over the 40D.

Ray, thanks for all your contributions here.  If MBroad has had the fewest posts in the last four years (1) you have got to have the most (6003) in the last six years, as of this writing  

There probably isn't anyone on here who tries to offer more input than you, and I have to say, it was actually your repeated posts and comparisons of the 50D and the 40D that helped me finally "pull the trigger" and get the 50D today, and what you have articulated above is exactly the difference I was looking for. There may be 100 other people on here who have made similar comparisons, but only you seem to take the time to post them.

I have been vascillating and procrastinating on the 40D, 50D, and Nikon D300 for quite awhile, but after being turned off by the way the Nikon D300 feels in my hands compared to the 50D, your contributions on the difference between the 40D and 50D have cinched it.

I am not sure what this has to do with the D3x, LOL, but I just thought I would add my $0.02 here on your post above.

Happy (soon to be) New Year,

Jack



.

[Ray]

I have been vascillating and procrastinating on the 40D, 50D, and Nikon D300 for quite awhile, but after being turned off by the way the Nikon D300 feels in my hands compared to the 50D, your contributions on the difference between the 40D and 50D have cinched it.

I am not sure what this has to do with the D3x, LOL, but I just thought I would add my $0.02 here on your post above.

Happy (soon to be) New Year,

Jack

Glad to be of help, Jack. I try to post the sort of comparisons of equipment which I would like to see myself when undecided about the relative merits of different models and brands of cameras and lenses.

I would have no hesitation in recommending the 50D over the 40D.

What has this got to do with the D3X? I consider the comparisons that have been made between the 40D and 50D, by myself and by Emil and others, as being the sort of comparisons that need to be made between the D3X and D3, so we can get an idea of just how much image quality improvement the D3X has to offer for the high price. Fine words by themselves won't convince me of anything.

A happy new year to you too, and a prosperous one so you can afford to buy a good telephoto lens for that 50D   .

[Dan Wells]

Getting back to the D3x, I just printed 24x36 from mine yesterday. The conditions were far from optimal (although the image was carefully taken - tripod, ISO 100, 14-bit NEF) - the printer was not mine (used a Z3100 at my photo dealer), I don't have my big computer with me to postprocess on, and the paper was not my favorite Hahnemuhle, but rather whatever Calumet had in their HP. Even so, it produced an amazingly sharp image at 24x36 inches - a 25x enlargement! It is certainly capable of printing that big on a routine basis, and my next printer will certainly be a 24 inch model to accommodate that. I would say that I am slightly more comfortable printing 24x36 from the D3x than 16x24 from the 1Ds mkII (which means that the D3x is getting quite a bit more detail per pixel, because it has only 1.5x the pixels, but the print area is just over twice as large). That detail difference could be in the AA/Low Pass filter, which is probably weaker and possibly a new design in the D3x. The D3x also has significantly higher DR (I'd say quite a bit more than 1 stop, but I haven't seen any real numbers yet), and is visually noiseless even in the deepest shadows at base ISO (the 1Ds mkII is low-noise, but NOT noiseless in the shadows, even at base ISO). Overall, a performance I have NOT seen from any DSLR before, and the images on screen look like the one sample I was able to print was not a fluke - it really is that good (the files really do look medium-format - I would say the subjective film equivalent is significantly over 6x9, but not yet 4x5 - some of the MFDBs DO reach 4x5 quality).

                                                         -Dan

[Ray]

I notice that Imaging Resource has images of both the D3 and D3X on their Comparator. Playing around with such images it's easy to see that the D3X has about the same noise as the D3 at ISO 6400, when the D3X is downsampled to the D3 size.

In fact, the D3X noise appears to be slightly greater in the IR images I used for comparison, but the D3X image is also slightly sharper than the D3 image when downsampled.

Panopeeper should pay attention to this. Downsampling a D3X image to 12mp using bicubic sharper, results in an image that is noticeably sharper and more detailed than the D3 image, at ISO 6400. Downsampling using standard bicubic still produces a slightly sharper result than the D3 image.

This additional sharpness would allow one to reduce noise in the D3X image whilst still retaining at least equal resolution to the D3, at ISO 6400.

The first image below compares the downsampled D3X image using bicubic sharper. The second shows what I would describe as slightly greater noise in the black cup which has been lightened considerably in 'Levels'. In the second image I used standard bicubic for downsampling.

[attachment=10654:D3X_down...mparison.jpg]  [attachment=10655:D3X_down...ISO_6400.jpg]

Of course, it goes without saying, when the full resolution of the D3X is required for a large print, an upsampled D3 image just can't compete. The resolution difference is greater than that between the 40D and 50D, as one would expect. A doubling of pixel count always produces a worthwhile increase in resolution.

[BernardLanguillier]

I notice that Imaging Resource has images of both the D3 and D3X on their Comparator. Playing around with such images it's easy to see that the D3X has about the same noise as the D3 at ISO 6400, when the D3X is downsampled to the D3 size.

For what it is worth both Thom Hogan and Iliah Borg claim after having compared the D3 and D3x side to side that overall, they will keep using their D3 beyond a certain ISO because they like the overall result better (that must include noise, detail, DR, colors,...):

- Iliah says that the threashold is around ISO 800,
- Thom has not given a definite number yet

Cheers,
Bernard

[Panopeeper]

I notice that Imaging Resource has images of both the D3 and D3X on their Comparator. Playing around with such images it's easy to see that the D3X has about the same noise as the D3 at ISO 6400, when the D3X is downsampled to the D3 size.

I find it a worthless excercise to play the downsampling game with JPEG images (pre-sharpened and lossy).

Panopeeper should pay attention to this. Downsampling a D3X image to 12mp using bicubic sharper, results in an image that is noticeably sharper and more detailed than the D3 image, at ISO 6400. Downsampling using standard bicubic still produces a slightly sharper result than the D3 image

If Panopeeper downsizes an image with bicubic sharper, then the noise does not get "improved"; can YOU do that better? See Emil's condemnation regarding bicubic sharper.

This additional sharpness would allow one to reduce noise in the D3X image whilst still retaining at least equal resolution to the D3, at ISO 6400

BS

Forget the above:

Happy New Year

[Panopeeper]

For what it is worth both Thom Hogan and Iliah Borg claim after having compared the D3 and D3x side to side that overall, they will keep using their D3 beyond a certain ISO because they like the overall result better (that must include noise, detail, DR, colors,...):

- Iliah says that the threashold is around ISO 800,
- Thom has not given a definite number yet

It is not disputable, that the D3's noise is lower at ALL ISO's. Therefor it is the question of personal preference, and even more of the specific setting (lighting, subject, requirement) if the higher pixel count or the lower noise is important.

For example if I am shooting for a pano and it does not cause much headache in that setting, I may pick a longer lens, shoot more frames with the D3 and have the same resolution as with the D3X in less frames, while having enjoyed a higher dynamic range. In other situations the D3X could be the better choice.

IMO these cameras should not be compared as competitors.

Bernard,. you too: Happy New Year (here, at the Pacific coast, it is still several hours away).

[Ray]

For what it is worth both Thom Hogan and Iliah Borg claim after having compared the D3 and D3x side to side that overall, they will keep using their D3 beyond a certain ISO because they like the overall result better (that must include noise, detail, DR, colors,...):

- Iliah says that the threashold is around ISO 800,
- Thom has not given a definite number yet

Cheers,
Bernard

Then they should show us their results, describe their comparison methodology and tell us precisely what it is about the D3 images they prefer. I can draw conclusions only from what I see when the medium is visual.

[Ray]

I find it a worthless excercise to play the downsampling game with JPEG images (pre-sharpened and lossy).


If Panopeeper downsizes an image with bicubic sharper, then the noise does not get "improved"; can YOU do that better? See Emil's condemnation regarding bicubic sharper.


BS

Forget the above:

Happy New Year

Maybe. However, if your point is valid it raises at least three questions. Why are Imaging Resource producing worthless images for comparison? Why should the lossy jpeg compression favour the D3X? Where are the RAW images from these cameras, of identical scenes, that could provide a more accurate impression of noise and resolution?

Happy New year to you too   .

[Panopeeper]

Why are Imaging Resource producing worthless images for comparison?

I did not say they are worthless for comparison. I said downsizing of JPEGs and comparing the result is worthless. If you want to downsize the D3X images, then pls in lossless form, without output sharpening.

Where are the RAW images from these cameras, of identical scenes, that could provide a more accurate impression of noise and resolution?

On the Thumbnail page. However, the D3X images can be converted only with Nikon Capture at the moment. Michael got a "pre-production copy" (a beta version) of ACR for his test.

I can measure the noise with Rawnalyze, but that is a single aspect of the comparison.

[BernardLanguillier]

Bernard,. you too: Happy New Year (here, at the Pacific coast, it is still several hours away).

Happy new year to you as well, all the best for 2009!  

Cheers,
Bernard

[BernardLanguillier]

Then they should show us their results, describe their comparison methodology and tell us precisely what it is about the D3 images they prefer. I can draw conclusions only from what I see when the medium is visual.

You could also get both bodies. Thinking of yourself as a high end consultant, you should be able to make 1500 US$ per day easily, considering the time you are spending at LL, you should be able to buy a D3 and a D3x within 2 weeks.

Cheers,
Bernard

[BernardLanguillier]

On the Thumbnail page. However, the D3X images can be converted only with Nikon Capture at the moment. Michael got a "pre-production copy" (a beta version) of ACR for his test.

Raw Developper 1.8.2 does also support the D3x if you are on OSX.

Cheers,
Bernard

[Ray]

I did not say they are worthless for comparison. I said downsizing of JPEGs and comparing the result is worthless. If you want to downsize the D3X images, then pls in lossless form, without output sharpening.

But you already know, Gabor, that I would never be so foolish as to compare different size images when addressing specific issues such as noise and resolution. Different size images are simply different images, different with respect to size, and therefore they require a different viewing distance for comparison purposes, which is very awkward.

Whether the source image is jpeg, tif or raw, whether it's been pre-sharpened or not, the image sizes have to be equalised for meaningful comparison from the perspective of the photographer who is concerned with a print or display output.

The question still remains, why does the processing that Imaging Resource has applied to images in their 'Comparator', favour the D3X over the D3 when the D3X image is downsized?

On the Thumbnail page. However, the D3X images can be converted only with Nikon Capture at the moment. Michael got a "pre-production copy" (a beta version) of ACR for his test.

It will be interesting to see how significantly the result changes when the same procedure is applied to unsharpened RAW conversions.