5DsR/MPE65 useful magnification experiments

[dlg]

First serious post here. I got the new Canon 5DsR, 50 MP full frame sensor :-) I posted this on Canonrumors, but thought some of you here may also be interested.

I've been playing around with it to see how it behaves in the macro area. I shot some small orchid flowers (Oberonia cf. lunata), total height 2 mm each, with MPE 65 mm at 1:1, 2:1, 3:1, 4:1, 5:1 and with TC 1.4xIII at 7:1. Each set at f/2.8 and f/4: all open for maximum resolution (here meant as the traditional separating two points in object), and 1 stop down for possible aberration correction and overall improved image quality. Between 60 and 175 frames were shot per set-up. Flashed manually with MT-24Ex at around 1/64-1/32 power, so very short exposure times. Flash heads mounted off lens to ensure consistent illumination angles. Stacking done on Cogynsis Stackshot with steps down to 19 µm. Stacking steps calculated as 70% of depth of field with circle of confusion = 0.03 mm in 8 x 10" print.

Processing on 6 core MacPro soup can with 32 GB RAM. Mac OS Yosemite does not reliably display thumbnails of CR2 file icons in finder. No idea why. RAW files were run through DxO (latest download). It takes time to process >100 files, but can be done on MacPro. Activity monitor had all processors going full throttle for several minutes, with quite a bit of heat coming out of the vent. I would not advise doing that on a laptop.

In previous tests with 5D2 files, RAW file processing was quicker in DxO than in PS CS5.5 extended with batch processing.

Stacking in ZereneStacker with 300 MB 16 bit .tif files was flawless. For me the P-max algorithm works best.

With the huge file sizes and small pixels (4.16 µm), the question arises at what magnification is nothing gained anymore in terms of better information? This point is reached at 4:1, where f/2.8 becomes effective f-stop f/14. This is a bit higher than what diffraction limited calculations arrive at (f/6.7–11, in most discussions). f/4 resulted in slightly softer images.
Comparing 5D2 images of same plant to the 5DsR, 5D2 still gains information at 5:1 (did not try 7:1), but still on fewer pixels. With 5DsR you can take a larger field of view at 4:1, print larger, and crop heavier, and get same information, mostly as expected.

Rik (pers. e-mail) commented that f/14 cut-off is largely in line with some other experiments posted/discussed here http://www.photomacrography.net/forum/v ... 164#101164 The extrapolated cut-off for FF 50 MP is f/17. I think the discrepancy lies in Rik shooting test-charts with B/W line pairs, while I shoot a real-world object with much lower subject/object contrast so lower starting point at MTF, and also in orange tones, so a about 10% longer wave length than standard 550 nm. 1 f-stop lower for real world in orange sounds about right.

I did shoot the same plant also at 5:1 with effective f-stop f/16.8, and I could not get any more information out of the file.

Attached is a 4:1 f/2.8 image, 111 image stack: full image height down sampled to 1000 pixels, cropped some of the black side areas out. Then second image is a 100% crop at 1000 pixels wide Again, flowers are 2 mm high, the bubbles are individual cells, about 20–40 µm in diameter, resolution limit seems to be at around 10 µm, which is about right for anything short of epi objective lenses on compound microscope. On dedicated stereomicroscope you can get down to around 4–6 µm (1.22 lamda/NA). The image is a straight output from Zerene. No cleaning, no capture sharpening, no output sharpening, just the straight stacked file.

Have to dig out the SEM image of a similar species, just for comparison. Some of the people here may get a kick out of it.

[rjlittlefield]

dlg, welcome aboard here at photomacrography.net.

For the benefit of other readers, let me provide some further analysis that was contained in that personal e-mail you mentioned.

BTW, I noticed in your posting [at CanonRumors] some discussion as follows:

the question arises at what magnification is nothing gained anymore in terms of better information? This point is reached at 4:1, where f/2.8 becomes effective f-stop f/14. This is a bit higher than what diffraction limited calculations arrive at (f/6.7–11, in most discussions). f/4 resulted in slightly softer images.

You may be pleased to hear that your observation is consistent with appropriate calculations.

At f/14, the spatial cutoff frequency for green light, 550 nm wavelength, is 130 cycles per mm. At the absolute minimum of 2 samples per cycle, this requires a pixel pitch of 3.85 microns, very much in line with the 4.16 microns pitch of a 50MP fullframe sensor. Any wider aperture will definitely out-resolve the sensor.

But I agree that you'll often see numbers like f/6.7-11 quoted in discussions. Those numbers come from sources such as http://www.cambridgeincolour.com/tutori ... aphy-2.htm, where if you plug in 35 mm full frame and 50 megapixels, you get that "Diffraction may become visible" at f/6.2 and "Diffraction Limits Standard Grayscale Resolution at f/9.4". The difference between those numbers and your f/14 is due to the fact that they're asking "At what aperture does diffraction begin to make an already degraded image look worse?", and you're asking "At what aperture does the sensor start to capture all the detail that's present in the optical image?" Different questions give different answers, even though the physics of diffraction is the same.

This is discussed in further detail in the thread surrounding http://www.photomacrography.net/forum/v ... 164#101164, which demonstrates by direct experiment with a test pattern that an f/11 optical image contains detail fine enough to require 47 megapixels in APS-C. Scale that to full frame and 50 megapixels, and the corresponding aperture comes out to be f/17, quite close to your observation with the MP-E 65 and your macro subject.

I don't recall anybody posting earlier about direct measurement of where empty magnification kicks in with an MP-E 65, so your test is very helpful at expanding the community's experience.

--Rik

[dlg]

dlg, welcome aboard here at photomacrography.net.

For the benefit of other readers, let me provide some further analysis that was contained in that personal e-mail you mentioned.

BTW, I noticed in your posting [at CanonRumors] some discussion as follows:

the question arises at what magnification is nothing gained anymore in terms of better information? This point is reached at 4:1, where f/2.8 becomes effective f-stop f/14. This is a bit higher than what diffraction limited calculations arrive at (f/6.7–11, in most discussions). f/4 resulted in slightly softer images.

You may be pleased to hear that your observation is consistent with appropriate calculations.

At f/14, the spatial cutoff frequency for green light, 550 nm wavelength, is 130 cycles per mm. At the absolute minimum of 2 samples per cycle, this requires a pixel pitch of 3.85 microns, very much in line with the 4.16 microns pitch of a 50MP fullframe sensor. Any wider aperture will definitely out-resolve the sensor.

But I agree that you'll often see numbers like f/6.7-11 quoted in discussions. Those numbers come from sources such as http://www.cambridgeincolour.com/tutori ... aphy-2.htm, where if you plug in 35 mm full frame and 50 megapixels, you get that "Diffraction may become visible" at f/6.2 and "Diffraction Limits Standard Grayscale Resolution at f/9.4". The difference between those numbers and your f/14 is due to the fact that they're asking "At what aperture does diffraction begin to make an already degraded image look worse?", and you're asking "At what aperture does the sensor start to capture all the detail that's present in the optical image?" Different questions give different answers, even though the physics of diffraction is the same.

This is discussed in further detail in the thread surrounding http://www.photomacrography.net/forum/v ... 164#101164, which demonstrates by direct experiment with a test pattern that an f/11 optical image contains detail fine enough to require 47 megapixels in APS-C. Scale that to full frame and 50 megapixels, and the corresponding aperture comes out to be f/17, quite close to your observation with the MP-E 65 and your macro subject.

I don't recall anybody posting earlier about direct measurement of where empty magnification kicks in with an MP-E 65, so your test is very helpful at expanding the community's experience.

--Rik

Hi Rik,

I did not want to include all the info you kindly provided by personal e-mail without asking for permission, so just paraphrased the punch-line. There is always more to it, no question. Your response and previous post/discussion is very illuminating. Thanks for pointing me to it.

[Peter M. Macdonald]

dlg,

This is a very interesting thread and provides food for thought. However, something appears to have gone wrong with the sizes of the two pictures which you have posted. They are supposed to be 1000 pixels on the long side. However, they are coming out at 850 x 768 pixels for the downsampled full frame image and at 768 pixels square for what was intended to be an actual pixel crop of an area 1000 pixels square.

Given the somewhat brutal way that bulletine board software changes the size of images, it may well be that neither your efforts nor the MPE 65mm are being displayed to their full potential.

Best,

Peter

[cactuspic]

Dig, I found your tests absolutely fascinating. As with the best of posts, it left me with both useful information and a pocketful of questions. Please forgive any imprecision in my questions, but I lack the scientific training and familiarity with optical formulas that many in this forum possess. First, is the point of empty magnification different for different sensor sizes and/or pixel densities? For example, would using the 65mpe on a 50 megapixel 5DR s yield the same result on a 24 megapixel 5D Mark III?

Second, my take on the article on diffraction (The Diffraction Limit, by Tim Parkin at https://www.onlandscape.co.uk/2012/08/t ... too-small/) is that the is a great deal of additional info that may be coaxed out of diffraction limited apertures with proper sharpening, particularly with high megapixel sensors. Is it possible that the empty magnification point would shift if both the 4x and 5x images were sharpened in accordance with the article?

Lastly, (and this may be more properly asked of Rik) would our final resulting image have greater resolution if we pulled out the latent detail in each of the constituent diffraction limited images of the stack by sharpening them prior to stacking?

[JohnKoerner]

Very interesting review, thank you.

My goal is to get the 5DSr by springtime. I don't typically shoot ultra-close (usually 0.5-2x:1).

I have heard, read, and seen that it is an incredible upgrade for Canon shooters, especially so for truly sharp macro lenses (Zeiss 100/Sigma 180).

Will bookmark this.

[rjlittlefield]

Lastly, (and this may be more properly asked of Rik) would our final resulting image have greater resolution if we pulled out the latent detail in each of the constituent diffraction limited images of the stack by sharpening them prior to stacking?

Sorry, I missed seeing this question until now.

For resolution, it doesn't much matter whether you sharpen before or after stacking. Locally, what comes out of the stacking process is pretty much exactly what went into it (plus accumulated noise if you're running PMax).

There can, however, be other benefits in sharpening first. Here is what I wrote for one of the Zerene Stacker FAQs:

Should I sharpen before or after running Zerene Stacker?

In many cases it doesn't matter. If you have a good sharpening tool that can identify and sharpen real detail while leaving pixel noise largely unchanged, then you may get better results by running it before Zerene Stacker. That's because it improves the “signal-to-noise ratio”, which helps Zerene Stacker to make better decisions. But with sharpening filters that do not have that level of discrimination, you'll get similar results either way and it's quicker and safer to just sharpen the final result. If you do sharpen before stacking, be sure not to oversharpen because fixing that problem would require reprocessing the whole stack.

The same reasoning applies to noise reduction. If you can reduce noise while preserving real detail, then that improves the signal-to-noise ratio and it's a good thing to do before running Zerene Stacker. Just don't overdo it, again because fixing that problem would require reprocessing the whole stack.

As a matter of practice, I'll sharpen and noise-reduce first if I'm going through Lightroom so that it's convenient to do that as part of the raw development process. But if I've shot a deep stack as in-camera JPEG then I'll defer all that to the very end.

--Rik

Edit: tweak the FAQ quote to match current wording.

[Asha]

Very nice--thanks for posting!

[Bushman.K]

To expand Rik's explanation just a bit, I'd say that in general case, sharpening with global methods is useless because it obviously amplifies both useful details and noise. Local sharpening, limited in some way, does its job better. Sharpening, applied after certain "smart" noise reduction also works for good.