Canon TS-E 24mm f/3.5L II - test images

[Bart_van_der_Wolf]

Last week, I had the opportunity to test the TSE 24, vs TSE-17 vs TSE-17 +1.4 TC.

Hi Michael,

I'm looking forward to the results!

Correct me if I'm wrong, Bart uses the 24 mm much for stitching with a panohead, while I'm interested to extend the FOV of a shot, called flatstitching and shifting in single shots.

I do both. Stitching with a panohead allows to exceed the FOV that delivers good quality with a flatstitch, but it's not something that suits all subjects. Flatstitching is relatively fast and is easy to post process, but it requires shooting with a horizontal displacement of camera+lens to avoid parallax (see my RRS solution to do that here).

Analog to what I saw in Bart's tests, a shift of about 7 mm keeps image quality to the required degree; while in the above example the chimney at the left side, beeing at 8 mm wouldn't fit it.

Yes, in the long dimension. You can shift further in the short dimension (e.g. horizontal shift with portrait orientation). For shooting I use the full +/- 12mm shift because some images have less detail towards the L/R edges but the 'breathing space' may be welcome. I use a simple piece of alumin(i)um to set the spacing of the index stop bars in the above mentions RRS setup. One can always crop the excess FOV afterwards (and there may be spatially variant sharpening tools to recover a few more millimetres of shift). The maximum amount of shift that delivers top quality results also depends on output size.

Shifting 5mm to both sides gives a image with a 1: 2-ratio, whith a HFOV of 90 degs, (equivalent to a 18 mm on a prime) while of course the VFOV remains stable.

Sometimes, I don't want to use a wider focal lengts (adding sky and grass) but want to have a long building - like this stadium - contextualised in the surrounding, here the skyscrapers. For that purpose, that TSE 24 works well, if shiftet about 7 mm only.

I agree, it's usually plenty wide. Going wider also introduces projection distortion (anamorphic distortion).

Did somebody can explain the distortion at the skyscraper under the "unshiftet" line, at the left? Its roof is not not horizontal, even if I set the verticals correct in post.

It probably is anamorphic distortion. One edge of the building is closer than the other and thus magnified more/less, but more importantly also 'distorted/stretched' by the projection on a flat plane. It's the stretching that dominates here.

Cheers,
Bart

 

[Michael Fontana]

I hadn't that much time for testing available, but here's the big picture:

- the 17 mm is better in the corners than the TSE 24, as the MTFs point out, as well. You can shift it further than the 24, while keeping image quality. As a con, the 17 mm is for my needs to wide; I would use rarely this lens; beside that I' ve the N 14-24, which fills my UW-needs. Flatstitching with the TSE 17 is aesthetically not really a option, while technically doable.

The TSE 24 fits much better my needs, for flatstitching as well, and yes, I have a similar setup (with stops as well) to avoid parallax, with the advantage of beeing light-weighted, compared to a nodalpoint stitch - setup. That rail would be in the bag all the time,when the TSE 24 will come with, offering multiple possibilties while avoiding a big bag. The stadion-shot in post 60 shows very well the flatstitch potential, to get the stadion contextualised, whithout using a wider angle, therefore making it smaller, from a given point - often you can't step back further. --> Flatstitching can save your day, whithout carrying much heavier gear.

Now the 17 mm+TC 1.4 - beeing a 24 mm as well, surprisingly produces quite sharp images until the corner; for a daily newspaper that combo would be fine. I disliked the noticeable degree of distortion and CA, which make it - for my needs - unusable.

I'll post some other samples, if I find the time.


Bart:
and there may be spatially variant sharpening tools to recover a few more millimetres of shift

Bart, you care to explain? When I tried that, I often get °crushed° details - even when using layermasks. I think that there is no substitutes for solid and healthy pixels - of course there is a marge to play arround, but I never managed to set that really far.

 

[Bart_van_der_Wolf]

I hadn't that much time for testing available, but here's the big picture:

- the 17 mm is better in the corners than the TSE 24, as the MTFs point out, as well. You can shift it further than the 24, while keeping image quality. As a con, the 17 mm is for my needs to wide; I would use rarely this lens; beside that I' ve the N 14-24, which fills my UW-needs. Flatstitching with the TSE 17 is aesthetically not really a option, while technically doable.

That's what I'd expect. Even single shifting a 17mm will risk significant anamorphic projection distortion. But in confined spaces, it can make the difference.

The TSE 24 fits much better my needs, for flatstitching as well, and yes, I have a similar setup (with stops as well) to avoid parallax, with the advantage of beeing light-weighted, compared to a nodalpoint stitch - setup. That rail would be in the bag all the time,when the TSE 24 will come with, offering multiple possibilties while avoiding a big bag. The stadion-shot in post 60 shows very well the flatstitch potential, to get the stadion contextualised, whithout using a wider angle, therefore making it smaller, from a given point - often you can't step back further. --> Flatstitching can save your day, whithout carrying much heavier gear.

That's the reason I first went for the TS-E 24mmL myself. It's very flexible, even more so with the right accessories for flatstitching. What's more, with 24mm the tilt is still a very useful addition. With a wider angle lens like 17mm or wider, the DOF will soon be adequate for getting a wide range of distances in focus. The wider the angle, the less the tilt weighs in, but the more the shift makes a difference. The 24mm is benefiting from both adjustments, and can be used with a relatively wide aperture, e.g. to stop motion, and still get the focus plane where it's needed.

Now the 17 mm+TC 1.4 - beeing a 24 mm as well, surprisingly produces quite sharp images until the corner; for a daily newspaper that combo would be fine. I disliked the noticeable degree of distortion and CA, which make it - for my needs - unusable.

That combination is said to have a similar sharpness as the 'old' TS-E 24mm. That wasn't adequate for my needs as well.

I'll post some other samples, if I find the time.

I know, it takes time to prepare a meaningfull comparison. Time is not always readily available to complete such a task.

Bart:
and there may be spatially variant sharpening tools to recover a few more millimetres of shift

Bart, you care to explain? When I tried that, I often get °crushed° details - even when using layermasks. I think that there is no substitutes for solid and healthy pixels - of course there is a marge to play arround, but I never managed to set that really far.

Sure. The distortion + loss of sharpness towards the corners is mathematically complex to characterize, and thus to correct. Normal sharpening methods apply the same correction across the entire image, they are called spatially invariant, IOW they do not vary across the image. More advanced sharpening can correct for a gradual sharpness fall-off towards the corners. That could already be called spatially variant. Even better will be a method that also (first) corrects distortion caused by the residual lens aberrations, and then sharpens the image. Even more advanced are methods that also address spatially variant blur due to motion, that is a real challenge to solve.

Fortunately, technology progresses and such advanced methods are becoming available for us mortals (not just in a CSI fantasy setting). Especially with powerful multicore computer systems, processing times are decreasing all the time, while the quality of the output is still improving.

I'm researching such stuff, and may be able to offer a software solution in the future to tackle the issue. Don't hold your breath just yet, I'm limited in my time and financial resources so it can take a while. Sponsors are welcome ;-)

Cheers,
Bart

 

[Michael Fontana]

Hi Bart

- any valuable informations or hints about Tilt with the TSE-24?

I used to know tilt it from 4/5' but not on DSLR, as the PC Schneider 28 hasn't it.
I saw quite a lot of °fun° shots with it, that I dind't find that interesting (Not yours)

- How does shift & stitch (NNP, with panohead) works?
Ex: camera in vert. position, having shiftet vertically for 5 mm; 3 images stiched....
Might be interesting for tall buildings...

Thanks for pointing me towards that lens, and for all the valuable informations, you provided: my copy of the TSE-24 will arrive at monday...

with the N1424, it'l be my dreamteam for archi

 

[Bart_van_der_Wolf]

Hi Bart

- any valuable informations or hints about Tilt with the TSE-24?

Hi Michael,

The wider the FOV, the less tilt degrees are needed. With a flat floor surface, and a camera height of around 1.5-1.7 metres above the ground plane (or from a wall if you shoot along a wall that needs to be in focus) you will only need something like 1 degree of tilt. Because I don't know how accurate the degree markings are on the lens barrel, I usually determine best focus with Live View, going back and forth between the near focus and far focus points. However, without Live View (and let's assume also without an extra precision groundglass, which gets a bit dark at f/3.5 but still allows to focus accurately in the viewfinder) you can use a table such as published by Keith Cooper:

Click on the table below to go to the author's webpage that explains its use:​

​

I used to know tilt it from 4/5' but not on DSLR, as the PC Schneider 28 hasn't it.
I saw quite a lot of °fun° shots with it, that I dind't find that interesting (Not yours)

Yes, the reverse tilt is a bit of a gimmick, giving the effect as if toys are being shot close-up, hence with limited DOF. It may be occasionally interesting to use it creatively (to throw the viewer's sense of proportions off balance), but again it's become a bit of a gimmick.

- How does shift & stitch (NNP, with panohead) works?
Ex: camera in vert. position, having shiftet vertically for 5 mm; 3 images stiched....
Might be interesting for tall buildings...

Shift
Not all Panostitchers handle shifted (or even decentered) lenses well. I think you use PTGui, and that probably allows to optimize for horizontal/vertical offset as well as PTAssembler that I use. I usually first optimize without these 'd' and 'e' offset parameters, then switch off all other parameters, and optimize alone for 'd' and/or 'e', then switch those off and reoptimize the other parameters. If you know the amount of shift in pixels, you can also input those values manually (1 mm in the 36mm dimension is 1/36 x # pixels in that dimension, and 1/24th in the other dimension). Things might get a bit tricky to calculate when you rotate the lens to an other angle between -90 and 90 degrees, so you might as well let the optimizer do it for you.

Tilt
Now here comes the tricky part. Tilting also shifts the entrance pupil in this lens! That means that your 360 degrees no-parallax rotation point moves in the tilt direction! Aye, what were the designers thinking, and it is not simple to calculate how much. So first see if you can use the DOF of the unshifted lens, thus avoiding the tilt. Next, try to stick to simple tilts at 90 degree angles + DOF, because you can still rotate quite good in one direction (perpendicular to the tilt direction). For instance, tilt down for a floor aligned focus plane, rotate horizontally through the vertical no-parallax axis for wide FOV. With modest tilt angles that will introduce very little parallax.

Shift + Tilt
I'm thinking about a method to determine either the needed entrance pupil adjustments before shooting, or as a second option correct for it afterwards but then a significant overlap is still needed to mask parallax between tiles.

Your specific Shift+Tilt scenario example is quite feasible. The camera is in portrait orientation, and is leveled and shifted up vertically to lose some of the foreground and gain height. The vertical axis of horizontal rotation (for a wider FOV) is still in the no-parallax position you determined earlier. Adding some tilt to position the focus plane from foreground to background (requires aligning the Shift and Tilt directions, which is easy with this lens) will hardly influence the alignment with the no-parallax axis for horizontal FOV. Only if you need to use a non-aligned tilt direction, you will introduce an offset that creates some parallax.

Thanks for pointing me towards that lens, and for all the valuable informations, you provided: my copy of the TSE-24 will arrive at monday...

You're welcome. Oh dear, that's going to be one long weekend ahead ;-)

with the N1424, it'l be my dreamteam for archi

Yes, pretty close to ideal. A dreamteam indeed.

Cheers,
Bart

 

[Michael Fontana]

Thanks Bart

studing the table from Keith, I got aware why you use Live View.
you need quite subtle hands to adjust 0.1 degs with that knob....

Shiftet stitch
Good to know - I forgotten - PTGui handles these params - 'e' for vertical shift resp. 'd' for the horiz. one.

With that function, the shiftet pano is possible, while avoiding the softwarebased °shift° and forced recalculation of the pixels, therefore the result should theoretically be superior. Did you took some tests of shiftet stitch vs 2-row-stitch?

Tilt
I will apply shift on most occasions - not that much the tilts, so I' ll use the DOF of the unshifted lens anyway.

Well, as long as the film plane (sensor) is vertical, and the lens tilts (but not vice versa) the lines will still be vertical.


Next, try to stick to simple tilts at 90 degree angles + DOF,
what do you intend by simple tilt? whithout shift?

With modest tilt angles that will introduce very little parallax.
might not work perfectly for small interiors, but as distance to the objects grow, could be ok for exteriors.

I think 1 deg of tilt is modest - that would provide at cam height of 135 cm the plan of focus identical with the ground....

°My specific scenario °
I believe it could work with shifts already, using a bit of hyperfocal distance - that would be the easier case. (The idea behind is to avoid softwarebased °shifts° with a fix lens and a two-row-scenario) A 24 mm has quite a large DOF and I don't believe that shifting is shortening it: for a Coc of 0.03 mm at a focus point of 2.5 m and f-stop of 8, you' re fine already..... from 123 till infinity....

Now, if you add vertically tilt to the shift, there's a risc of getting the base oof... whithout regaining it back while closing the f-stop..... it might be hard to control these subtilities in the viewer...

I fear a little bit, too, IQ like distortion might loose with shift and tilt, while beeing stable with one of thes only.
But off course, these are speculations at the moment...


that's going to be one long weekend ahead ;-)
no doubts :-

When flatstitching, you take all 3 shots, or just the left and right?
For the stadium-flatstitch I used the two, only

 

[Bart_van_der_Wolf]

Thanks Bart

studing the table from Keith, I got aware why you use Live View.
you need quite subtle hands to adjust 0.1 degs with that knob....

Yes, but due to the DOF of a 24mm lens used at normal distances, things are not that critical for normal size output. LiveView does help for significant enlargements.

Shiftet stitch
Good to know - I forgotten - PTGui handles these params - 'e' for vertical shift resp. 'd' for the horiz. one.

With that function, the shiftet pano is possible, while avoiding the softwarebased °shift° and forced recalculation of the pixels, therefore the result should theoretically be superior. Did you took some tests of shiftet stitch vs 2-row-stitch?

For multirow stitches I don't use shift. I prefer to shoot and use the best part of the image circle when I need to push the FOV beyond what can be done in a single row. The 24mm is long enough to allow high resolution stitched of close up subjects, without the need for huge numbers of tiles. Not pushing the shift too much will help to maintain that image quality towards the corners of a multilayer stitch. I think I did try doing multirow shifted images, but the optimization adds many more degrees of freedom to arrive at a solution, so the overall optimization could suffer. But the main reason to do unshifted multirows is due to maximizing image quality.

In certain situations it can help image quality to shoot shifted multirow tiles instead of relying on interpolation, e.g. when shooting something overhead from close-by (such as a detail of a facade in a narrow street) IOW where significant distortion needs to be corrected. In such a case it's better to capture real pixels than calculating them.

Tilt
I will apply shift on most occasions - not that much the tilts, so I' ll use the DOF of the unshifted lens anyway.

Well, as long as the film plane (sensor) is vertical, and the lens tilts (but not vice versa) the lines will still be vertical.

Yes, perspective control is the most important issue with wide angles. Tilt can help for interiors, or when the composition has important foreground detail, but also distant detail at the same time.

Next, try to stick to simple tilts at 90 degree angles + DOF,
what do you intend by simple tilt? whithout shift?

No, I meant that it's easier to stick to the rotation clickstops (at 45 degree intervals), because they can be reproduced and corrections can be pre-calculated. The lens allows to rotate the shift and tilt directions to intermediate positions as well. An example of using tilt at an intermediate angle is when e.g. shooting a facade at an angle (whith shift to keep straight verticals). Not only is there foreground , but the background/facade is also at different focus distances from left to right. One can use a 45 degree rotation of the tilt angle, getting a bit of both tilts needed. The benefit of that is that it's still possible to use relatively wide apertures, e.g. to reduce subject motion or camera shake on an unstable floor. Of course with extreme shifts one needs to use a narrower aperture to improve image circle edge quality.

°My specific scenario °
I believe it could work with shifts already, using a bit of hyperfocal distance - that would be the easier case. (The idea behind is to avoid softwarebased °shifts° with a fix lens and a two-row-scenario) A 24 mm has quite a large DOF and I don't believe that shifting is shortening it: for a Coc of 0.03 mm at a focus point of 2.5 m and f-stop of 8, you' re fine already..... from 123 till infinity....

Now, if you add vertically tilt to the shift, there's a risc of getting the base oof... whithout regaining it back while closing the f-stop..... it might be hard to control these subtilities in the viewer...

Yes, but it depends on the distance, the space one has to work in. The DOF in my experience, when combined with just a bit of tilt, allows huge DOF including the base of the building. But then there are different scenarios as well, e.g. some close overhead detail, or shooting through a gate or door opening. Different scenarios require diffrent approaches, but the versatility of the lens is not the limiting factor. You already have the experience of working with adjustments, so the learning curve is not that steep anymore.

When flatstitching, you take all 3 shots, or just the left and right?
For the stadium-flatstitch I used the two, only

When shifting in the long dimension direction, only 2 shots are needed, with enough overlap to align them manually. When you shift in the short dimension direction, there is theoretically zero overlap on a 24mm height full frame (-12 + +12 =24 mm), so either don't shift the full 12mm in each direction, or shoot a third image at zero offset.

Cheers,
Bart

 

[Michael Fontana]

Bart

I got that lens that afternoon and had just a bit of time for the first tests; build is fine, there's even a lock of the tilt-mechanism, and the knobs can be replaced by bigger ones.

As the first test, I just went on the road, and played a bit arround, took some photos from the houses here, to verify how much the shift-mechanism brings in real life; indeed - shifting some mm on a 24 mm brings much more than shifts with a 28 mm.
So for my needs, the TSE-24 is wide enough. I noticed again the nearly complet absence of distortions.

... Not pushing the shift too much will help to maintain that image quality towards the corners of a multilayer stitch

I agree completly about not shifting when doing multirow stitches, for IQ reasons -
I had been thinking rather on a easy-going single-row-stitch, example: a house, with a few mm shiftet upwards to get roof and sky into the frame - to avoid a multirow stitch.

In certain situations it can help image quality to shoot shifted multirow tiles instead of relying on interpolation, e.g. when shooting something overhead from close-by (such as a detail of a facade in a narrow street) IOW where significant distortion needs to be corrected. In such a case it's better to capture real pixels than calculating them.

I agree again, when the software has to correct to heavy, we will have pixel degredation.

Thanks, I can see now what you meant by simple tilts; a 45 deg. of tilt can cover °two directions° - well, creating a diagonal plane of focus in the space.

I found the curvature of field of that lens quite big - in some further tests I had to stop down to 11 to avoid image unsharpness due to curvature of field:
here's the full image, with a shift of 7 mm, the focus had been at the windows with the lamp. It's clearly not a to short DOF, as some bricks, with the same distance from the lens, are in focus:

and here's a 100%-screenshot in the RAWconverter - lower right side of the image:

The trade-off is between stopping down the f-stop to correct the curvature of field vs diffraction.

 

[Michael Fontana]

You can send the 3 shots for flatstitches - taken whithout the correcting rail - to PTGui, it will corrcet; the math errors (cp-distances) are so low, due to low distortions. So Basically this reduces the time for making to 5 min...
Works even with architecture ;-)

ok, it will probably not work with close interriors - I haven't tried that - but good to know that it's ok with a distance of some metres...

 

[Asher Kelman]

Tilt
Now here comes the tricky part. Tilting also shifts the entrance pupil in this lens! That means that your 360 degrees no-parallax rotation point moves in the tilt direction! Aye, what were the designers thinking, and it is not simple to calculate how much. So first see if you can use the DOF of the unshifted lens, thus avoiding the tilt. Next, try to stick to simple tilts at 90 degree angles + DOF, because you can still rotate quite good in one direction (perpendicular to the tilt direction). For instance, tilt down for a floor aligned focus plane, rotate horizontally through the vertical no-parallax axis for wide FOV. With modest tilt angles that will introduce very little parallax.

Bart,

Do you mean use a little forward tilt of the camera in portrait position and then rotate on the Z axis?

Asher

 

[Bart_van_der_Wolf]

Do you mean use a little forward tilt of the camera in portrait position and then rotate on the Z axis?

Yes, that's correct. The tilted plane of focus doesn't have to be foreground - background, or closer left side - further away right side (or vice versa), it can be rotated at any angle, with clickstops at 45 degrees for a diagonal plane of focus. Together with DOF that should cover most needs.

Cheers,
Bart

 

[Bart_van_der_Wolf]

I found the curvature of field of that lens quite big - in some further tests I had to stop down to 11 to avoid image unsharpness due to curvature of field:

Yes, the sweet spot seems to be at f/11 or f/13, where the corner sharpness becomes more similar to the center. The corners need some help to reduce the aberrations, and the center gets hit by some diffraction. Surprisingly f/16 is still usable (with subsequent sharpening) on a 1Ds3 with its 6.4 micron sensel pitch. This is actually very good for a 24mm, and miles better than its 24mm predecessor.

here's the full image, with a shift of 7 mm, the focus had been at the windows with the lamp. It's clearly not a to short DOF, as some bricks, with the same distance from the lens, are in focus:

I'm not fully decided about it being curvature of field with my lens (I have to assume yours in no different). If that's the case then refocussing the corners would improve things there. You can also watch the plane of focus intersecting e.g. a brick or gravel road when you shoot wide open. I don't see much deviation from a straight intersection line, but there is deterioration towards the corners/edges when shifted, let's call it residual aberrations (probably a combination of several aberration types). There is also the possibility that the click stops for tilt give an ever so slightly non parallel alignment between the principal planes of the lens and the sensor. They are all accurate within some mechanical tolerance. It's probably easiest to test for it with tethered shooting, which allows a large preview.

The trade-off is between stopping down the f-stop to correct the curvature of field vs diffraction.

You are finding the same things I did. See my earlier corner crops at full shift. Chromatic aberration is almost non-existing, geometrical distortion is very minimal, hardly noticeable at all, and the lens handles high contrasts (windows with a view) without complaining. Shifting has its limits if the absolute best quality is needed, but a little shift goes along way already on a 24mm. Also, not all subjects need corner to corner focus (e.g. sky is not critical), so sometimes one can choose a camera position to exploit that (one can even dial in a bit of foreground or edge blur by tilting in the opposite direction or at 90 degrees).

Once you get a feel for where the limitations are (mostly shift related), you'll start to like the lens even better. Right now you are searching for the limitations, so you will find some. In practice they don't bother me as much anymore. Try printing, and be amazed. And for that very last bit of quality, there is still the option to stitch, but you'll be doing it with possibly the sharpest 24mm on the planet, and it shows.

I'm looking forward to your impresssions as the lens grows a bit on you. One needs to get familiar with the possibilities and peculiarities, and there are many of the former, and fewer of the latter. One of the things I like is that close up detail is so incredibly sharp, so that allows to exploit the perspective in a composition by placing something interesting at close range. Tilting will then allow to keep the focus throughout the image, but due to the distance the background draws less attention. Good sharpening allows to recover some of the losses from closing the aperture to f/11 - f/16 (e.g. Capture One allows to correct for sharpness fall-off).

Enjoy the lens, and play with it a lot, it will help to speed up the process of learning its characteristics. I almost exclusively shot with the lens for non-critical work during 2 weeks, and it became second nature to use it. I even use it in combination with the high precision focus screen that's intended for f/2.8 or wider lenses. The f/3.5 means the viewfinder gets a bit darker than with the standard screen, but it's still easy to focus most subjects.

Bart

 

[Michael Fontana]

Thanks Bart

your hint, using e in PTGui worked perfectly, stitches from 3 images and a panohead (not flatstitch):

The first image is unshiftet - the 2nd shiftet by 5 mm only - what a difference!!
The ceiling is much clearer in the 2nd one - its even a bit exagerated, 3mm might have been apropriate. IQ is still very good in the stitch with 3 mm shift.

Some 24 mm are hard to stitch, but the low distortion of the TSE helps a lot to make easy stitching.

 

[Bart_van_der_Wolf]

Thanks Bart

your hint, using e in PTGui worked perfectly, stitches from 3 images and a panohead (not flatstitch):
...
The first image is unshiftet - the 2nd shiftet by 5 mm only - what a difference!!
The ceiling is much clearer in the 2nd one - its even a bit exagerated, 3mm might have been apropriate. IQ is still very good in the stitch with 3 mm shift.

Yes, you've got it figured out! It's the best of both worlds, portrait orientation stitching of a few horizontal tiles, with a bit of vertical shift (often not much shift is needed). The result is super resolution and a wide FOV adapted to the subject, not restricted by the sensor aspect ratio.

This may be one of the best uses of the combined techniques. The Stitching software needs to accommodate shifted lenses (to avoid another preprocessing step and keeping records about the shift amounts used), which PTGui and PTAssembler do.

Cheers,
Bart

 

[Michael Fontana]

Yes Bart

what you wrote makes much sense - especially for my archistuff - a easy 3-frame shift-stitch (with 73 degs in height!!) is not that much more of work than a single shot while offering many advantages. That makes the lens more versatile, again...

And yes the jump from 28 mm shift to the 24 shift is quite bigger than from 28 prime to 24 prime, cause every mm of shift has a bigger influence, it's like opening another door.

I'm aware that I' ve to play a bit with that lens, to know it better - I'm sure on a sunny day, it will show differently than with today's clouds.

I'm not fully decided about it being curvature of field with my lens (I have to assume yours in no different). If that's the case then refocussing the corners would improve things there. You can also watch the plane of focus intersecting e.g. a brick or gravel road when you shoot wide open. I don't see much deviation from a straight intersection line, but there is deterioration towards the corners/edges when shifted, let's call it residual aberrations (probably a combination of several aberration types)

I' m aware that the newer Canon lenses have a different curvature of field (way more speheric) than the one I use most (Zeiss) so I might go back and study that one - maybe a bit of backfocusing could do it. Off course, with shifting - end even more when tilting - focusing require's a new attention, which is fine - it gives back a forgotten tool.

I better go now and have dinner...

 

[Michael Fontana]

I struggled yesterday in lowlight-situations with focusing, hence my questions:

What's the nature of focus on a shiftet lens?
Of course it's different than 4/5', as on a TSE not all glas elements move, unlike tje lens on a bellow cams.

Bart, when shifting, do you use the focus position of the unshitfet lens - or you focus, when shiftet?

These questions might rise as well for tilting, but lets go through shifting first.

Edit: when stitching, a 24 mm is rather wide, at the right and left side of the stitch we have distorstions (in linear projektion), due to the wide angle of 100 degs, even when pan 24 degs only. Therefore reducing the HFOV in the stitch makes sense, for a flat surface, like a facade....

 

[Asher Kelman]

On anamorphic Distortion: where does it occur in the 24mm v 24 mm T-S lens?

Edit: when stitching, a 24 mm is rather wide, at the right and left side of the stitch we have distorstions (in linear projektion), due to the wide angle of 100 degs, even when pan 24 degs only. Therefore reducing the HFOV in the stitch makes sense, for a flat surface, like a facade....

Michael and Bart,

I think, if I'm following correctly, that this gets us back to anamorphic distortion that presents at the lateral margins of the regular Canon 24mm lens. (The double base and its musician gets very fat, LOL!). Is there software to correct such anamorphic distortion, (maybe DXO?)? Then, what about PTGUi, Auto Pano Giga and other stitching software?

Take the case of using the unshifted 24mm T-S lens to make a pano by rotating the lens through its entrance pupil (/nodal point), Is it your sense that one should crop the images before stitching when using the 24 mm T-S or is that anamorphic distortion in this larger lens is now located further out, (in the currently unused), lateral margins of the larger MF image circle of the T-S lens?

Asher

 

[Bart_van_der_Wolf]

I think, if I'm following correctly, that this gets us back to anamorphic distortion that presents at the lateral margins of the regular Canon 24mm lens.

Hi Asher,

The anamorphotic distortion is the same for all rectilinear lenses on flat sensors. That distortion is only due to the projection on a flat plane, and in the case of wide angles of view the projection angle gets very oblique. That's why I often refer to it as projection distortion, since it has little to do with the lens (other than covering a wide field). BTW the distortion is only apparent when we view the image from the 'wrong' viewing distance. When we e.g. magnify the image 8x and we view the image from 8x the focal length, everything is okay (other than that some may need reading glasses ;-)). It is because we usually look at a wide angle image from too far away that we get the impression of distortion towards the edges/corners. The shorter the focal length, or the wider our stitched FOV, the more this can become an issue.

Is there software to correct such anamorphic distortion, (maybe DXO?)? Then, what about PTGUi, Auto Pano Giga and other stitching software?

DxO has that functionality built-in, and it seems to do a good job. Because I use PTAssembler (Windows) a lot, I know from personal experience that it allows to choose from several interactive methods to mitigate the 'effect'.

Take the case of using the unshifted 24mm T-S lens to make a pano by rotating the lens through its entrance pupil (/nodal point), Is it your sense that one should crop the images before stitching when using the 24 mm T-S or is that anamorphic distortion in this larger lens is now located further out, (in the currently unused), lateral margins of the larger MF image circle of the T-S lens?

Again, the effect is purely dependent on the (in this case chosen) projection method. When we choose rectilinear (as if we had a larger flat sensor projection), then the effect will be exactly the same as a lens that has the same FOV. Because we are hardly limited to a fixed FOV with stitching, it tends to be a common occurrence, hence the provisions in PTAssembler. In general it is best to limit the FOV to less than 120 degrees anyway because FOVs wider than 100 degrees are almost impossible to adjust convincingly, especially on architectural subjects or scenes involving people. So that would set a 14mm lens with its 104 degrees FOV at the limit of what's usable. Any adjustment to reduce the projection 'distortion' will introduce distortions itself (straight lines/diagonals can get warped).

Cheers,
Bart

 

[Asher Kelman]

Hi Asher,

The anamorphotic distortion is the same for all rectilinear lenses on flat sensors. That distortion is only due to the projection on a flat plane, and in the case of wide angles of view the projection angle gets very oblique. That's why I often refer to it as projection distortion, since it has little to do with the lens (other than covering a wide field). BTW the distortion is only apparent when we view the image from the 'wrong' viewing distance. When we e.g. magnify the image 8x and we view the image from 8x the focal length, everything is okay (other than that some may need reading glasses ;-)). It is because we usually look at a wide angle image from too far away that we get the impression of distortion towards the edges/corners. The shorter the focal length, or the wider our stitched FOV, the more this can become an issue.

So for the picture of the orchestra with the 24 mm it should normalize at ~10 inches wide viewed from 8 inches, which is the perfect distance for my uncorrected vision. I'll try that, LOL!

DxO has that functionality built-in, and it seems to do a good job. Because I use PTAssembler (Windows) a lot, I know from personal experience that it allows to choose from several interactive methods to mitigate the 'effect'.

Can one shrink the lateral parts of the image in width to normalize circles? If so, one can add that as a corrected layer to substitute one or two people at each side.

Again, the effect is purely dependent on the (in this case chosen) projection method. When we choose rectilinear (as if we had a larger flat sensor projection), then the effect will be exactly the same as a lens that has the same FOV. Because we are hardly limited to a fixed FOV with stitching, it tends to be a common occurrence, hence the provisions in PTAssembler.

So what are the provisions in PTAssembler, I'll check it out.

In general it is best to limit the FOV to less than 120 degrees anyway because FOVs wider than 100 degrees are almost impossible to adjust convincingly, especially on architectural subjects or scenes involving people. So that would set a 14mm lens with its 104 degrees FOV at the limit of what's usable. Any adjustment to reduce the projection 'distortion' will introduce distortions itself (straight lines/diagonals can get warped).

Thanks for the reminder on the boundary of ~ 105 degrees, (14mm) for a reality boundary.

Asher

 

[Bart_van_der_Wolf]

So for the picture of the orchestra with the 24 mm it should normalize at ~10 inches wide viewed from 8 inches, which is the perfect distance for my uncorrected vision. I'll try that, LOL!

10 inches wide is 7x 36mm wide, so 7x 24mm focal length is 6.7 inches viewing distance.

Can one shrink the lateral parts of the image in width to normalize circles? If so, one can add that as a corrected layer to substitute one or two people at each side.

When one shrinks in one dimension, then circles become ovals. Ovals may become circles, depending on where they are in the image.

So what are the provisions in PTAssembler, I'll check it out.

Here are the projections offered: http://www.tawbaware.com/projections.htm

The 'Recti-Perspective' projection can be used if vertical lines and radial lines (passing through the center) must remain straight. The horizon will be the only straight horizontal line.
The 'Squeezed Rectilinear' projection allows to maintain a rectilinear projection in the center, but start compression beyond a given FOV.
The 'Recti-Cylindrical' projection combines rectilinear projection in the center and Cylindrical projection towards the sides, thus keeping verticals straight.

Those are a few of the projections that can also be tweaked, but there is also a possibiity to bake one's own, aptly called 'user defined', or make a combination of 4 different projections in 4 quadrants of the image (not all projections combine to a meaningful composition though).

Thanks for the reminder on the boundary of ~ 105 degrees, (14mm) for a reality boundary.

It never hurts to keep reality in view, especially when stitching removes so many physical boundaries. Architecture however restricts our degrees of freedom, because usually straight lines need to be kept straight.

Cheers,
Bart

 

[Asher Kelman]

10 inches wide is 7x 36mm wide, so 7x 24mm focal length is 6.7 inches viewing distance.



When one shrinks in one dimension, then circles become ovals. Ovals may become circles, depending on where they are in the image.



Here are the projections offered: http://www.tawbaware.com/projections.htm

The 'Recti-Perspective' projection can be used if vertical lines and radial lines (passing through the center) must remain straight. The horizon will be the only straight horizontal line.
The 'Squeezed Rectilinear' projection allows to maintain a rectilinear projection in the center, but start compression beyond a given FOV.
The 'Recti-Cylindrical' projection combines rectilinear projection in the center and Cylindrical projection towards the sides, thus keeping verticals straight.

Those are a few of the projections that can also be tweaked, but there is also a possibiity to bake one's own, aptly called 'user defined', or make a combination of 4 different projections in 4 quadrants of the image (not all projections combine to a meaningful composition though).

Bart,

You are so helpful and Max Lyons is an excellent writer and programmer to be able to mix various projections. The solutions are something I dreamed about!! I wish AutoPano Giga could do this but it doesn't as yet. If there are no Mac programs that can do what PTassembler does, I can set it up in Parallels on my Macbook Pro. It seems exactty what's needed for large panos where one wants the architecture to seem right and then to be able to add people on a separate layer taken with the correct angel of perspective and distance but not otherwise corrected.

Asher

 

[Michael Fontana]

As Bart already replied Asher's question, I' might add only, that PTGui - a panorama stitcher we discussed here too, has more projections than APP, including compression as well, l did upload and link it here at OPF, a while ago:
QT-mov

 

[Asher Kelman]

As Bart already replied Asher's question, I' might add only, that PTGui - a panorama stitcher we discussed here too, has more projections than APP, including compression as well, l did upload and link it here at OPF, a while ago:
QT-mov

That movie is very helpful, Michael!

What did you use to record it? I think now I'll upgrade Parallells and get PTgui on to my MacBook Pro! It seems we have to go to the "Dark Side" to benefit from these incredible advances.

Another program to note for altering the width of things doesn't rely on perspective and projection, but actually removes marked objects or uses fractals to rebuild added structure when dimensions are increased. It's part of CS4 but I've not played with it yet.

Asher

 

[Cem_Usakligil]

That movie is very helpful, Michael!

What did you use to record it? I think now I'll upgrade Parallells and get PTgui on to my MacBook Pro! It seems we have to go to the "Dark Side" to benefit from these incredible advances.

Another program to note for altering the width of things doesn't rely on perspective and projection, but actually removes marked objects or uses fractals to rebuild added structure when dimensions are increased. It's part of CS4 but I've not played with it yet.

Asher

Encouraged by Bart's kind help and advice, I have ended up buying the PTAssembler this week, next to the PTGui Pro which I already had. Both programs are similar but are quite different in particular areas. Especially, the number of available projection methods in PTAssembler is mind boggling and the level of control offered to the user is next to none (mind you, like Bart I am running the newest beta version of PTA which is very stable). There are some hybrid projection methods which combine 2-4 methods to get the best results in different parts of the picture. I have managed to correct the elongated/slanted distortion of people's heads at the edge of a 17mm shot quite easily by manually tweaking the projection parameters. Also, I take a lot of hand held HDR brackets. PTA makes it very easy to align the multiple shots and export them to aligned files which then can be fused by TuFuse or Enfuse or any other blending or tone mapping software. All said, I am very enthusiastic about it.

Cheers,

 

[Michael Fontana]

Asher

the tool for recording is called screenflick -
PTGui runs nativly on Mac. You can use it for reducing anamorphotic distortion as shown in the movie for a pano and even better on a single image as well, I sometimes use it on the 14 mm of the 14-24, see example below.

It's done in a small margin with the slider, or for constant results better within the script, here with a 0.2-correction on the horizontal axis, meanwhile the vertical is left to 0:

Here's the code:
#-rect_compression_x 0.2
#-rect_compression_y 0