SL2's software correction of M lenses?

[bags27]

Advertisement (gone after registration)

Erwin Puts's latest blog entry is below [I added the embolden, as that is the main point]. When I passed this on in a PM to Tailwagger, with whom I've been having a running conversation on matters SL2, M10, and M glass, he produced some photos to demonstrate what might confirm Puts. I've urged him to post his results, so I've started this thread, since to the extent and how M glass is compatible with the SL2 seems to be an important question for many (including me).

It's becoming accepted that M glass is more compatible with the SL2 than with any non-M camera (by any maker). Sean Reid has pretty much demonstrated this, and Jono Slack has pretty much said the same. But is it only hardware at work here? Puts is suggesting it's at least partly in the software and that's what Tailwagger set out to investigate. 

Something strange is going on! I'm slowly processing the results of my exploration of the Leica SL2 and the Summilux-SL and I'm encountering strange results. The first, less remarkable result, is the fact that on the SL2 the Summilux-SL and the Summilux-M lens perform practically the same, at least in the middle and on the test cards I use. What is remarkable, however, is the performance at the different apertures. Already at full aperture both lenses perform almost as well as at the central diaphragms where normally the best performance can be expected. The Summilux-M is recognized by the software of the SL2, because this lens already has the electronic identification. The suspicion is now strong that the software in the camera increases the performance. With the M8.2, the performance of the SX-M is as expected: slow increase from full aperture to f/5.6 and then a decrease to f/16.
The many recent and super bright lenses from Canon, Nikon and Panasonic (from 0.95 to 1.4) also show remarkably good results at full aperture in test shots. Despite the many lens elements and the larger diameter of the bayonet mount, these results seem to be a bit over the top. Because these modern lenses only fit on digital and proprietary brand cameras, it is impossible to determine whether and to what extent the software actually helps the lens designers. It cannot be ruled out and questions the so-called lens tests of the magazines and the many bloggers.

https://photo.imx.nl/blog/

[Tailwagger]

1 minute ago, bags27 said:

When I passed this on in a PM to Tailwagger, with whom I've been having a running conversation on matters SL2, M10, and M glass, he produced some photos to demonstrate what might confirm Puts. I've urged him to post his results, so I've started this thread, since to the extent and how M glass is compatible with the SL2 seems to be an important question for many (including me).

When Ken mentioned the Puts article to me, It seemed as though this would be a relatively easy thing to test. For example, if one had a cheapo adapter which didnt communicate the codes along with the Leica adapter which does, one could compare the results to see how much and what sort of correction was being applied.  But as I only have the M Adapter L, I thought perhaps I could perhaps tape over the lens codes. But that seemed a little messy and potentially dangerous. It turned out be a little tricky to fool the camera as whenever I mounted my coded Summilux 50mm, the camera recognized immediately recognized it and automatically applied the lens profile. 

The trick, in the end was to:

1. Turn the camera off.
2. Mount the M adapter L as normal
3. Mount the lens but only partially so that the code cant be read by the sensor
4. When the menu for lens selection appears,  click unknown and turn corrections off
5. Fully mount the lens.

This enabled me to shoot with and without corrections.   I'll state up front that I'm not a test wonk and this particular test was not well controlled. ie. the results prove nothing. I was just curious and decided to quickly see if there was anything at all to support Puts notions. The shots were hand held. With the lens being unknown to the camera, the test shot was taken at 1/30" (unknown) and 1/40" (known) respectively (interesting in and of itself as there they were only a few seconds apart indoors with no perceivable lighting change). As I have IBIS on by default, in hindsight, its possible therefore any perceived difference in acuity could be down to camera shake as when the lens is unknown IBIS is turned off. If there's sufficient interest I'll repeat (or others can) the test on a tripod in slight more controlled conditions.

All that out of the way here's what I sent to Ken. Both shots were taken at f1.4 ISO 100.

Full frame without correction:

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

With corrections:

Again not on a tripod so framing shifts slightly.  What is not immediately evident in the above given the compressed jpgs is the difference in color balance. The raw uncorrected image in LR is warmer with a slight yellow cast, the corrected one is cooler with a comparatively bluish cast when viewed side by side.  100% crop next.   

 

[Tailwagger]

I focused using zoom on the number 8. 

Without correction:

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Now with correction:

Again, it possible that IBIS made the difference or perhaps I just nailed focus slightly better the second time around. Either way, Ken felt this was curious enough that we move the conversation into the public forum.  I will say that anecdotally I have been a little disappointed with the performance of my Zeiss Distagon. One wonders if there's some additional software tricks being applied to improve the performance of Leica lenses beyond simply making the cover glass thinner. 

[SrMi]

Sean Reid wrote that Leica SL/SL2 cameras, when recognizing M-lenses, apply corrections for color drift and luminance vignetting.

I am not sure which deficiencies Erwin Puts expected to see when wide open. He seems upset that in-camera software is used to improve the quality of the image.

 

[scott kirkpatrick]

The in-camera processing that was developed for the M series is of course available for the SL series (but apparently not used by other consortium members such as the S1 and S1R).  Sean  Reid also devotes great energy to being able to eliminate post-processing tricks such as distortion and LCA correction which the SL and TL/CL apply to their zoom lenses and wide angles, but the M lenses do not use this.  I've looked the files generated for M lenses by both M and SL cameras, and that set of corrections are just not used there.  So what is Puts seeing?  He looks at Siemens stars in the center of the frame, and says it's as good as it gets when either SL or M lenses are wide open, while in the past lenses could be expected to improve for a few stops, then soften due to diffraction.  Well, the SL Summicrons are designed to be diffraction-limited, at their best wide open.   And the only real problems with M lenses on the ultra high resolution sensors show up at the edges, so I'm not sure there is a problem.

[lct]

Did you try to remove the opcodes and compare with and without them? As far as distortion is concerned, differences can be pretty obvious on the CL.

[digitalfx]

Advertisement (gone after registration)

Im not sure I understand...why is this a problem?

It seems like it's a huge benefit...especially for those of us that also own M systems.

[Tailwagger]

9 minutes ago, digitalfx said:

Im not sure I understand...why is this a problem?

It seems like it's a huge benefit...especially for those of us that also own M systems.

 I don't think any one said that it was a problem. It's simply, at least from my perspective, extremely interesting. I think we would all expect vignette and distortion correction as a result of lens recognition. While the color correction was unexpected, at least by me, it's not particularly surprising that Leica might apply a profile. In fact, I think I now recall color corrections being discussed in other threads a while back.

The thing I found intriguing and something I would not have suspected is that somehow the image result is sharper when the camera recognizes the lens. So the question I had was whether or not I might see what Puts was on about.  And if so, further wondering just what the heck they are the doing to achieve this? Just applying an appropriate sharpening value for a given optic to the raw file (perhaps slightly concerning) or something more sophisticated, say, modeling? One of the first things that came to mind upon seeing the result was Leica's recently announced commitment to computational photography. Might this be a simple example of it?

24 minutes ago, lct said:

Did you try to remove the opcodes and compare with and without them? As far as distortion is concerned, differences can be pretty obvious on the CL.

I just followed the procedure I enumerated above.  Essentially removing the lens codes by rotating the lens to a point where it couldnt be read when the camera detected that the M-adapter-L was attached on startup.

[lct]

I meant removing the opcodes through DNG Cleaner.

[scott kirkpatrick]

DNG Cleaner removes a little section called WarpRectilinear from the DNG file.  That section does not appear in files generated by M cameras (including the M10M) so there is nothing for it to do. 

The L mount cameras do include WarpRectilinear opcodes in their DNG output.  The parameters can be set to do nothing at all.  They describe the mapping of the space that is outside into the captured image with a polynomial with four terms, and give these coefficients separately for each color R, G and B determined  at each pixel after demosaicing.  Having (slightly) different transformations for each color permits correcting for color focus discrepancies, or lateral color aberration.  When each color is transformed with parameters (1 0 0 0) the captured image is identical to the image outside, or undistorted.  That's what an SL or SL2 writes for an SL Summicron or for an M lens.

[scott kirkpatrick]

Here's an example of my M21 SE 3.4 @ f/3.4 showing what it can do on the SL2 without an y software assistance:

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

U1000912 by scott kirkpatrick, on Flickr

you can click through to Flicker (click twice) and see a very slight softening in the lower right corner, but nothing that would make me discard the image.

[lct]

4 hours ago, scott kirkpatrick said:

DNG Cleaner removes a little section called WarpRectilinear from the DNG file.  That section does not appear in files generated by M cameras (including the M10M) so there is nothing for it to do. 

The L mount cameras do include WarpRectilinear opcodes in their DNG output.  The parameters can be set to do nothing at all.  They describe the mapping of the space that is outside into the captured image with a polynomial with four terms, and give these coefficients separately for each color R, G and B determined  at each pixel after demosaicing.  Having (slightly) different transformations for each color permits correcting for color focus discrepancies, or lateral color aberration.  When each color is transformed with parameters (1 0 0 0) the captured image is identical to the image outside, or undistorted.  That's what an SL or SL2 writes for an SL Summicron or for an M lens.

Thank you Scott but those opcodes correct distortion don't they. Pretty obvious on the CL with M lenses but i have no experience on SL/SL2 cameras.

[jaapv]

Lateral colour aberration is something quite different from distortion.

[Chaemono]

vor 19 Minuten schrieb jaapv:

Lateral colour aberration is something quite different from distortion.

And it’s baked into the RAWs, not corrected by opcodes. As stated earlier, there aren’t any opcodes with SL2 files when using M lenses. 

[lct]

When using a coded wide M lens on your SL2, don't you see differences distortion wise if you remove the opcodes from DNG files really? Just curious as i do see such differences out of the CL.

[Chaemono]

vor 10 Stunden schrieb Tailwagger:

 I don't think any one said that it was a problem. It's simply, at least from my perspective, extremely interesting. I think we would all expect vignette and distortion correction as a result of lens recognition. While the color correction was unexpected, at least by me, it's not particularly surprising that Leica might apply a profile. In fact, I think I now recall color corrections being discussed in other threads a while back.

They need to do more of it for the 75 Noctilux in the next FW upgrade, IMO.  Lateral color aberration can be pretty bad with this lens on the SL2 in certain situations.

vor 3 Minuten schrieb lct:

When using a coded wide M lens on your SL2, don't you see differences distortion wise if you remove the opcodes from DNG files really?
[...]

No, I just did.  Only when I choose the lens profile in LR I see a difference in distortion. 

[jaapv]

Lateral CA is so simple to remove in postprocessing...

[Tailwagger]

9 hours ago, lct said:

I meant removing the opcodes through DNG Cleaner.

ahh, apologies, I see.... no, don't have a copy of the software.  

[scott kirkpatrick]

6 hours ago, jaapv said:

Lateral colour aberration is something quite different from distortion.

Actually it is not.  It is a distortion that is different for different wavelengths.  Here's a definition from the web:

Lateral Chromatic Aberration. Lateral color is an off-axis aberration resulting from a difference of image scale at each wavelength. Lateral color causes colored fringes at the edge of the field.

And one of the interesting things about the WarpRectilinear parameters is that they transform the image differently for each color to correct this.  The different colors' parameters typically differ by only a percent or so of the overall transformations. Note that this "opcode" is applied after the image is demosaiced and has R, G, and B values ready to be applied to each pixel postion.

[Advertisement]

Hallo,

Look here: Leica L-Mount

[scott kirkpatrick]

Here's a source of confusion.  I didn't realize that the CL, the SL and the SL2 put different opcodes into the output DNG.  This may change as the SL2 firmware gets brought up to the state that the product team eventually wants to reach, but at present the SL2 only generates distortion corrections for the L-mount zooms but prints default values for M lenses and the Summicrons, which have no effect.

Here's what the WarpRectilinear opcode actually contains in a file made with a CL using the CL35 Summilux.  The following lines were extracted from the DNG file:

<<<<

LensSpecificationExif: 35.0 mm f/1.0

*** Warning: Exif IFD LensMakeExif is not NULL terminated ***

LensMakeExif: "LEICA CAMERA AG"

LensModelExif: "Summilux-TL 1:1.4/35 ASPH."

 

Opcode: WarpRectilinear, minVersion = 1.3.0.0, flags = 0

Planes: 3

  Optical center:

    h = 0.500000

    v = 0.500000

  Plane 0:

    Radial params:     1.000710, -0.015868, 0.004349, 0.001293

    Tangential params: 0.000000, 0.000000

  Plane 1:

    Radial params:     1.000077, -0.015829, 0.004921, 0.000961

    Tangential params: 0.000000, 0.000000

  Plane 2:

    Radial params:     0.999930, -0.015013, 0.004551, 0.001189

    Tangential params: 0.000000, 0.000000

>>>>

The only lines that really matter are the three lines, one for each color "plane" labelled Radial Params.  If they were 1 0 0 0 that describes having no distortion, and needing no correction.  These coefficients are pretty small, and the differences between them (the LCA correction) are even smaller (around a per cent). 

Here are the opcodes (I took out more of the irrelevant stuff) for the M 21 Super-Elmar 3.4, shot on a CL:

<<<<

21 SE3.4 in an M to L Leica adapter:

LensSpecificationExif: 21.0 mm f/3.5

LensModelExif: "Super-Elmar-M 1:3.4/21 ASPH."

Opcode: WarpRectilinear, minVersion = 1.3.0.0, flags = 0

Planes: 3

  Plane 0:

    Radial params:     1.000403, -0.023597, 0.011575, -0.000911

  Plane 1:

    Radial params:     1.000048, -0.023363, 0.011650, -0.000956

  Plane 2:

    Radial params:     0.999815, -0.022986, 0.011723, -0.001012

>>>>

Those  params would be replaced by 1 0 0 0's in the SL2's output at present.

Note_  the SL also puts this information in, but the numbers are different because the sensor is bigger.

Edited January 19, 2020 by scott kirkpatrick