24-90mm Focus Shift (Diglloyd)

[cpclee]

Advertisement (gone after registration)

DSLR lenses simply can't be designed the same way because your viewing system is still entirely optical and hence the camera can't try to fix distortion digitally. That constrains the design of the lens and is not to be taken as an advantage -- though I'm quite sure you would!

 

If you look at tashley's default DNG out of the SL vs his own attempt at correcting distortion, you see that the camera can do a far better job (extremely evident in the periphery of the images) because (1) Leica knows the exact parameters of the optical design (2) Leica can fix things on a higher mathematical order than any packages like LR will allow you to.  The photog simply will have no way in correcting to the same exacting standards.

 

Systems that are not thus constrained are now all being designed with full use of optical and digital corrections wherever appropriate.  That includes Sony E, Olympus OM-D, Leica SL, as well as fixed lens cameras like Sony's RX1, Leica Q, Leica X, etc.

 

Images taken with the 24-90 are super sharp into the corners at all aperture and all distances. Somewhere on Erwin Puts' website he remarks how the 24-90/2.8-4 basically reaches the performance level of S system lenses.

 

I have yet to see any CaNikon wides that don't produce smeary corners.  It is the same with M wides.  They do far better than the CaNikon wides, but still with some noticeable deterioration towards the edges.  In this aspect the SL might be by far the best in any full frame system camera.

 

 

I use all my Canon lenses on a Sony A7R2, even if all of them were designed for their matching Canon bodies. They all shine one the 42 MP sensor, and the AF is working fine. I even get correct EXIF data.

Luckily, Canon engineers are not taking shortcuts (yet).

[scott kirkpatrick]

I suspect RawDigger gave you a file that has been down-ressed in order to fit the ugly corners into approximately 24 MPx.  

scott

 

Well I had that wrong. My longer essay on this is clearer.

 

scott

[scott kirkpatrick]

Sure, one could make it any size merely by up-resing. The question is, if you correct the distortion in the raw RAW file to the same degree as Leica's profile tells Lightroom to achieve and then crop the image so no white borders or black corners are showing, what pixel dimensions do you get without then doing the upresing step? I still think it's about 19mp. 

 

Here, Tim and I were using shorthand to say the same thing in different ways.  I say that there's is nothing magical about Leica's profile -- it's facts that only the lens designer can tell us (the distortion curve) wrapped in a procedure that Adobe lays out for anyone to follow.  The boundary in the distortion-corrected virtual image that contains 24 MPx maps, as Tim found, into the central 19 MP of the original raw data,  but I find that there is usable stuff that draws on more of the original raw data.  You just have to go out to more than 24 MP in the virtual image to see it all.

 

And, as Jared points out, the transformation in use is effective only at the outer parts of the image.  It's not up-ressing, which applies a uniform expansion across the whole image.

 

scott

[tashley]

OK, let's try to put this to bed - especially since we all seem to be largely agreeing as to the facts, and can decide the importance of the implications of those facts as individuals!

 

@ Scott, I agree that the C1 rendition with distortion corrections off is very similar to the RawDigger rendition - but it isn't quite the same. For example, it is rendering the file as 6000x4000 rather than 6016 x 4016. I have no idea what else it might or might not be doing, but that's precisely why RawDigger is so useful - because you know it isn't doing anything: it merely gives you the sensor readout and then you can choose how you want it displayed (RGB, RAW Composite, RAW per channel etc) but there are no options to confuse what is being rendered. 

 

However we can agree that if you export a TIFF from RawDigger and a TIFF from C1 with corrections off, they are very close. For practical purposes C1 isn't perfect but it's close enough.

 

We can also agree that if you want to salvage more useable pixels from the file in C1, you can do so by choosing to remove some or all distortion correction, and/or by accepting some vignetting and/or by changing the aspect ratio.

 

We can also agree that if the manufacturer's profile is applied, C1 will suggest a crop to match what the manufacturer wants us to see. That crop removes quite a bit of the frame. However, unless there's an option I am not seeing, when you export this as a TIFF, C1 automatically upscales it to 6000 x 4000 which is 24mp.

 

We can also agree that in applying the distortion correction, using Manufacturer's Profile, the peripheries of the image are pushed around by more than the centre since the peripheries are where the greatest correction is needed in order to achieve rectilinearity - if that's what you want (and it is what Leica is in effect suggesting). But I would observe that there IS data being stretched in the centre, and that if you flip at 100% view between corrected and uncorrected versions, you can easily see that. It's just less so in the centre.

 

However, my original claim, to which I (rather boringly!) stick, is that if you take the original sensor data from RawDigger (but here I will use C1 because it is so close and because it relates to your exposition above) and then process it so as to get the same (or as close to the same as is possible with available tools) result as the OOC JPEG, which is almost identical to the DNG file as rendered in LR, Leica's bundled editor (which applies Leica's corrections by default and they can't be switched off) OR in C1 with Manufacturer's Profile applied, you are end up with a 24mp image which is upresed from an approximately 19mp image.

 

These images make this fairly clear:

 

That  (above) is the DNG opened in C1 with lens corrections OFF and we can see that it suffers from distortion and severe vignetting. It is 6000x4000 =24mp

 

 

That (above) is the same file but with Manufacturer's Profile corrections ON. The cropped area (brighter area) is what will get output as a 24mp file. Clearly it is a crop: of the original data off the sensor, shown in the first of the above images, 24mp native, it is chucking a bunch of information away and squidging the rest of it around so as to achieve rectilinearity. In other words, at this focal length, the output as DNG or as OOC JPEG will be taking data from less than the 24mp native area of the sensor and up-resing it back to 24mp. Yes, a lot of the data is there, yes, the areas most affected are the peripheries and yes the result still looks very nice - but at 24mm focal length, it's a crop sensor camera/lens combo

 

{NOTE, added after reading Scott's post above this one, which he posted while I was writing this one: I use the loose shorthand 'up-resing' to refer to a more complex process, which includes some combination of up-resing and interpolation. It's sloppy of me but quicker.. my point is that data is being created in some places to make the file 24mp again}

 

So my original post was merely attempting to answer Jared's original point which was to question the quantum of the process.

 

In doing that, I was stuck with the unavailability of a perfect tool. C1 won't (or if it will I can't see how!) let me export that bright area in the image above at it's actual pixel size: it automatically reses it up to 24mp again. LR won't let me turn lens corrections off. So what I did was to take the raw RAW from RawDigger, export as a TIFF, use manual lens corrections in LR on that file so as to best approximate the Leica result and then crop it to as closely match the Leica result as possible. And I got a touch under 19mp. Others are welcome to try the same thing but with different tools but I will surprised if the result is very significantly different - I'm always willing to learn! But one glance at the images above shows you that it has to be about right because a whole bunch of what gets captured is then being jettisoned in order to get the same result as Leica gets from the file.

 

@ Ikarus, I think we are also in violent agreement. Somewhere above I gave some pixel dimensions and implied print sizes that matter to me and you have provided some that are relevant to you. Yours are at an assumed printing resolution of 300dpi and mine were at 180. Obviously when there's enough data, I like to print at 300DPI or even, a la Ming Thein, at much higher resolutions for an 'ultra print'. But really quite often I have an image that needs to go as large as possible and I have found that with my workflow I can 'get away with' as low as 180dpi and still have a really nice print, if the file is good. So I guess my point is that we all have different sized pinheads and therefore each of us needs different numbers of angels to fill them. With a 6000 wide really good quality file I can print to 6000/180 = 33.33 inches, handy since my printer is 24" and therefore this fits nicely onto a 24 x 36" print. 

 

But if the 6000 pixel wide file I am starting from is actually itself in effect already up-resed from, in my approximation a 5314 wide file, then really I wouldn't want to print it to more than 29.5 inches because 180 dpi is already quite a stretch.

 

That's a loss, to me and for my workflow, of 3.8" of printing width. It really is no biggie but it is something I need to know - and it's not nothing! And you're totally correct that the answer is to use a prime. But one does need to know that because the 24-90mm, with its size and price, might be taken by some to weigh and cost so much because it had a more prime-like performance than cheaper, lighter zooms. Indeed at many parts of its zoom range I am sure that is true - it gives really impressive results and is a wonderful thing.

 

@ Jared -  I totally agree that knowing more about how the MTF is produced would be useful.

 

Phew. The dog wants a walk...

[scott kirkpatrick]

I think you're wrong in saying "We can also agree that if the manufacturer's profile is applied, C1 will suggest a crop to match what the manufacturer wants us to see. That crop removes quite a bit of the frame. However, unless there's an option I am not seeing, when you export this as a TIFF, C1 automatically upscales it to 6000 x 4000 which is 24mp."

 

That crop, which you show in your second frame grab illustration, already has 24 MPx (because Adobe sez so).  No further upscaling.  It has already reached that pixel size through the transformation that uses the DNG parameters -- automatically in LR, or when you choose the 100% option in C1.

 

scott

[jrp]

The more important question than resolution is "Does the distortion-corrected picture have the field of view of a 24mm lens or is the effective focal length shorter 26mm, say?"

 

Is the preview on the back of the camera or viewfinder correct, or do you end up with a trimmed image?

[tashley]

Advertisement (gone after registration)

The more important question than resolution is "Does the distortion-corrected picture have the field of view of a 24mm lens or is the effective focal length shorter 26mm, say?"

 

Is the preview on the back of the camera or viewfinder correct, or do you end up with a trimmed image?

 

I can't answer that for sure but I'd have a strongish guess that the lens is actually a bit wider than 24mm so after the crop and corrections you're getting a 24mm FOV. I seem to remember someone had discovered that on the Q. 

[tashley]

I think you're wrong in saying "We can also agree that if the manufacturer's profile is applied, C1 will suggest a crop to match what the manufacturer wants us to see. That crop removes quite a bit of the frame. However, unless there's an option I am not seeing, when you export this as a TIFF, C1 automatically upscales it to 6000 x 4000 which is 24mp."

 

That crop, which you show in your second frame grab illustration, already has 24 MPx (because Adobe sez so).  No further upscaling.  It has already reached that pixel size through the transformation that uses the DNG parameters -- automatically in LR, or when you choose the 100% option in C1.

 

scott

Crikey this is complicated! I keep on thinking I must be missing something.

 

Can we agree that in the second of my C1 screen grabs above, the pixel data in the grayed out area of the frame isn't making it into the DNG? If so, let's call that number of discarded pixels "X"

 

From the original 24mp captured, X is discarded.

 

24-X<24   surely?

 

So whenever any of the raw processing programs tells us is that 24 is the total number of mp in the finished image, represented by the area which is not grayed out, it has to have gone from  24-X (which is less than 24) all the way back to 24. By any name, that is an up-res even if technically it is part interpolation during the application of the transofrmation. Or am I crazy and I just don't know it (possible!)?

 

Whether 24-X=19 is another question and there are unknowns (to me at least) in the form of the following:

 

The second screen grab shows the edges bowing inwards as a result of the application of the transformation. That means that pixels are being 'pushed in' rather than 'pulled out and having the gaps filled'. So the question is, in outputting 24mp is the process 'throwing away' the excess pixels that have been squeezed in, or is it leaving them there to take their rightful place as the image expands to fill 24mp again? I really don't know and I doubt of it's even that simple.

 

But in any event, we can clearly see that the areas that are greyed out are discarded. So I'd bet my last dollar that 24-X<24 but I can't prove that 24-X=19 and in fact that was only presented as an estimate. So my question is, Scott, what do you think 24-X is?

 

Whatever its exact quantum, to mis-quote Monty Python's parrot sketch, it is an ex-X. It is no more. It is deceased!

 

 

[adli]

I did a test to replicate the symptom. It seems to me as the focus error goes beyond the movement of the cross hair from 1x to magnified view.

 

I made a sheet filled with rectangles,with bold bars in the middle so I knew where focus was. Then I shoot it from a low angle the reproduce the effect. As you can see from my pictures, the focus cross hair moves down approximately one row when magnification is on, but the actual focus point seems to be  4-5 rows below the position of the cross hair at 1x.

 

Attached pictures:

 -Picture taken with Leica SL, focused on the bold cross in the middle of the picture, AFc.

 -Screenshot of LCD at 1x

 -Screenshot of LCD at 10x

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!

[cpclee]

1. Because in Leica's current implementation the crosshair doesn't enlarge proportionally with the magnification, you shouldn't really trust the size and position of it.  Use the crosshair only when magnification is turned off.

 

2. Looking at your screenshot of the LCD at 1x, I'm pretty certain the point of actual focus, which as you said looks like 4 - 5 rows down from the crosshair, is still within the focus area rectangle that is not showing in your screenshot.  The crosshair is towards the lower edge of the rectangle but still within the rectangle by my estimation.  

 

I commented on these two idiosyncrasies a few pages back

[jaapv]

Crikey this is complicated! I keep on thinking I must be missing something.

 

Can we agree that in the second of my C1 screen grabs above, the pixel data in the grayed out area of the frame isn't making it into the DNG? If so, let's call that number of discarded pixels "X"

 

From the original 24mp captured, X is discarded.

 

24-X<24   surely?

 

So whenever any of the raw processing programs tells us is that 24 is the total number of mp in the finished image, represented by the area which is not grayed out, it has to have gone from  24-X (which is less than 24) all the way back to 24. By any name, that is an up-res even if technically it is part interpolation during the application of the transofrmation. Or am I crazy and I just don't know it (possible!)?

 

Whether 24-X=19 is another question and there are unknowns (to me at least) in the form of the following:

 

The second screen grab shows the edges bowing inwards as a result of the application of the transformation. That means that pixels are being 'pushed in' rather than 'pulled out and having the gaps filled'. So the question is, in outputting 24mp is the process 'throwing away' the excess pixels that have been squeezed in, or is it leaving them there to take their rightful place as the image expands to fill 24mp again? I really don't know and I doubt of it's even that simple.

 

But in any event, we can clearly see that the areas that are greyed out are discarded. So I'd bet my last dollar that 24-X<24 but I can't prove that 24-X=19 and in fact that was only presented as an estimate. So my question is, Scott, what do you think 24-X is?

 

Whatever its exact quantum, to mis-quote Monty Python's parrot sketch, it is an ex-X. It is no more. It is deceased!

 

 

Are we sure that the sensor area is exactly 6000x4000? These are suspisciously round numbers.Could it  be that the actual active sensor area is larger to accomodate for the crop, like on the Q ? (And does the crop apply always, even if another lens is mounted - that would show up on Rawdigger as well?)

[tashley]

Are we sure that the sensor area is exactly 6000x4000? These are suspisciously round numbers.Could it  be that the actual active sensor area is larger to accomodate for the crop, like on the Q ? (And does the crop apply always, even if another lens is mounted - that would show up on Rawdigger as well?)

 

Jaap, according to RawDigger it is 6016 by 4016 so not significant enough to to account for most of X... As to the second part of the question I can't answer because my SL is long gone and I didn't have any adaptors to try other  lenses - sorry!

[adli]

 

Not sure what you mean? Are you trying in a complicated way to say that the camera focuses on the nearest point inside an imaginary rectangle around the cross hair?

 

 

1. Because in Leica's current implementation the crosshair doesn't enlarge proportionally with the magnification, you shouldn't really trust the size and position of it. Use the crosshair only when magnification is turned off.

 

2. Looking at your screenshot of the LCD at 1x, I'm pretty certain the point of actual focus, which as you said looks like 4 - 5 rows down from the crosshair, is still within the focus area rectangle that is not showing in your screenshot. The crosshair is towards the lower edge of the rectangle but still within the rectangle by my estimation.

 

I commented on these two idiosyncrasies a few pages back

[CheshireCat]

DSLR lenses simply can't be designed the same way because your viewing system is still entirely optical and hence the camera can't try to fix distortion digitally. That constrains the design of the lens and is not to be taken as an advantage -- though I'm quite sure you would!

 

 

Sure, and I never said I am against digital corrections.

The problem here is that I don't see what is the advantage of such an unbalanced digital/optical design in the case of the 24-90. The lens is huge, heavy, complicated, expensive, and a "throw-away" in case one switches to other camera systems.

I do enable digital distortion correction on some shots taken with - say - the ZE 21/2.8, but it is often not required and I leave it off for better image crispness (especially in the center). I can also use that lens on other systems, which is really important to me.

 

I have yet to see any CaNikon wides that don't produce smeary corners.  It is the same with M wides.  They do far better than the CaNikon wides, but still with some noticeable deterioration towards the edges.  In this aspect the SL might be by far the best in any full frame system camera.

 

So you are saying that the only advantage of this unbalanced digital/optical design is that it has better performance in the corner than any other lens on CaNikon and any other lens on M.

But I find it hard to believe that this "monster" is any better than - say - an Elmar-M 24/3.8 (on a M240) or a Loxia 21/2.8 on a Sony A7R2... and even if it is slightly better, two A7R2 cameras with a few excellent compact primes are more compact and less expensive than a single SL with this lens.

Now, if the 24-90 were way smaller and lighter, I would not be so negative.

 

[Peter H]

...........

Now, if the 24-90 were way smaller and lighter, I would not be so negative.

 

 

 

 

 

 

I expect that if they had designed the lens to have the same specifications but achieve no distortion or other imperfections that can be dealt with by digital correction, it would be even larger. And heavier and more expensive.

 

The benefit would be a few (how subjective of me!) extra pixels around the edge at the widest setting, and perhaps transportability to non-Leica bodies.

 

Otherwise, if the benefits were more substantial and the costs not so great, why wouldn't they have done it?

[CheshireCat]

It would be really interesting to understand if Sony plays the same digital tricks with their new G master lenses.

An interesting field test of their new 24-70 here:

http://briansmith.com/field-test-sony-g-master-24-70-f2-8-85-f1-4-lenses/

 

At the tele setting, Sony is 70/2.8 on 42MP while Leica is 90/4 on 24MP.

[CheshireCat]

I expect that if they had designed the lens to have the same specifications but achieve no distortion or other imperfections that can be dealt with by digital correction, it would be even larger. And heavier and more expensive.

[...]

Otherwise, if the benefits were more substantial and the costs not so great, why wouldn't they have done it?

 

Let's wait and see how the new Sony 24-70/2.8 (better specs on paper, at less than half the price) compares to this lens.

But I am afraid that the answer to your question is that even the best people sometimes make mistakes, especially when dealing with new technologies (for them).

[thighslapper]

Arne's tests exactly replicate mine ..... the crosshair is incorrectly placed ..... but despite this the maximum focus point still lies in front of point you think you are focussing on. 

 

The DOF puts the 'in focus area' almost completely in front of the users expected point of focus. 

 

If Arne does the same again with his camera upside down the crosshair error will partly correct this .... with the 'in focus area' moving back a bit ...... so there are two things going on ..... mismatch between crosshair and point of focus and the maximum focus point lying still further in front of that .....

 

.......... but the problem only seems to be apparent using test charts and close distances .... and appears to be absent in real-world use ......simple, isn't it  

[adli]

Same test with camera upside down (handheld this time).

 

It confirms what thightslapper predicted, partly corrected but still a bit off.

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!

[Advertisement]

Hallo,

Look here: Leica L-Mount

[Peter H]

Let's wait and see how the new Sony 24-70/2.8 (better specs on paper, at less than half the price) compares to this lens.

But I am afraid that the answer to your question is that even the best people sometimes make mistakes, especially when dealing with new technologies (for them).

 

 

There's no end of products with better on-paper specs than the Leica equivalent that sell at half the price or lower.  That's a different subject really; an important one, but different.

 

But am I understanding you correctly? Are you suggesting that the decision to use digital correction on the 24-90 was just a mistake?