Stitching by Shifting versus Twisting

[Doppler9000]

As the supply of the “unicorn” lenses dwindles, and prices rise, I was hoping to get some insights into peoples’ views on the two approaches to creating stitched images.

Jim Kasson investigated the question several years ago, using a Nikon SLR.

Slide or Spin?

While the equipment discussed here is different, the basic tradeoffs are the same.

A shifted mosaic is easier to reassemble in post, but relies on the lens’ peripheral performance.

An image that is made by rotating around the entrance pupil requires more sophisticated software, but uses the centers of the images.

Has anyone tested the approaches with any of the usual suspects?

 

[stevev]

Hi Doppler,

When you say, "tested the approaches with any of the usual suspects", do you mean tested by comparing flat stitching vs twisting/rotating stitching with the same lens, as I believe Jim did in his test?

I don't think he touched on what I find the best advantage of flat stitching, which is that visualising the final result is much easier, because apps like Viewfinder make it super simple. Also, I suspect a much better lens that the 24mm PC-E that he used would have yielded less chromatic aberration, better sharpness and less falloff than he experienced.

In any event, that sounds like an interesting test - I've not done it - however if the final result had similar IQ, I would still flat stitch (with a suitable camera & lens) to take advantage of bulletproof stitching, ability to tilt, ability to use rise/fall, easy visualisation, no adjusting the rail for pupil entrance on different lenses to deal with parallax error...

Steve.

 

[Doppler9000]

Hi Doppler,

When you say, "tested the approaches with any of the usual suspects", do you mean tested by comparing flat stitching vs twisting/rotating stitching with the same lens, as I believe Jim did in his test?

I don't think he touched on what I find the best advantage of flat stitching, which is that visualising the final result is much easier, because apps like Viewfinder make it super simple. Also, I suspect a much better lens that the 24mm PC-E that he used would have yielded less chromatic aberration, better sharpness and less falloff than he experienced.

In any event, that sounds like an interesting test - I've not done it - however if the final result had similar IQ, I would still flat stitch (with a suitable camera & lens) to take advantage of bulletproof stitching, ability to tilt, ability to use rise/fall, easy visualisation, no adjusting the rail for pupil entrance on different lenses to deal with parallax error...

Steve.

The test could use the same lens, or, for example, a readily-available lens with a smaller image circle versus a unicorn - something like a high quality 80mm or 90mm lens versus a 60 XL, for example.

“If the final result had similar IQ”. There are two parts to this - one, if the “if” is true, and two, can you find and afford the unicorn lens even if the IQ is similar?

 

[dchew]

I have done these comparison tests and prefer shifting by a large margin. However, I have the following situation:

1. I have most of those "unicorn" lenses I suspect you are referring to: sk60xl, sk100 (not the 90hr-sw but close), Rodi 138 and now the sk210T.
2. I don't have the desire/need to stitch together images with wide angles of view; I just don't shoot that wide. Jim K's test of stitching with a 24mm lens isn't relevant to my use.
3. I'm almost always doing single-row stitches to either a) make a panorama or b) give me a wider view with a specific lens by horizontal shifts with the back in portrait. Final format is 54x72 vs native 40x54.

I concur with what's already been said about the relative advantages / disadvantages of each approach. For me, the important advantages of flat stitching are:

• It is very quick. I can complete the image captures in just a few seconds.
• Long, flat horizons (like the ocean) don't get distorted at all by stitching software. I just use Photomerge in Photoshop and select "Reposition." That doesn't warp images.
• As Steve mentioned, you get to accurately see and assess the boarders of the finished image in-camera. There is no guessing whether everything will be covered correctly after the software does its thing.

The disadvantages, primarily image quality at the edges of the image circle used, don't apply to me because I specifically select lenses that can handle what I need. To me, this image circle evaluation is my most-valued criteria when making lens choices.

Dave

 

[Doppler9000]

Thanks, Dave. Joining the unicorn hunt, I suppose…

 

[dchew]

The disadvantages, primarily image quality at the edges of the image circle used, don't apply to me because I specifically select lenses that can handle what I need.

I shouldn't be so dismissive about this. Of course the image quality of any lens is lower as you move out from the image center. I just don't see it making a noticeable difference in prints or even on screen. I'm not shooting at f/5.6, and even with the 138 I'm often between f/8 and f/11. DoF, wind, focus accuracy and to a lesser extent diffraction are what I worry about more than image falloff from center.

Dave

 

[daz7]

For me the biggest issue with stitching by camera rotation is that you cannot apply all the camera movements freely (I am using a large format camera with digital). When shifting the back at the rear plane i have freedom to use all tilts, shifts and swings that I want or need. Also, there are no additional corrections needed and even free software will cope with the planar stitches without issues. Back stitching is the closest you can get to using a larger format films.

 

[Geoff]

I have done these comparison tests and prefer shifting by a large margin. However, I have the following situation:

1. I have most of those "unicorn" lenses I suspect you are referring to: sk60xl, sk100 (not the 90hr-sw but close), Rodi 138 and now the sk210T.
2. I don't have the desire/need to stitch together images with wide angles of view; I just don't shoot that wide. Jim K's test of stitching with a 24mm lens isn't relevant to my use.
3. I'm almost always doing single-row stitches to either a) make a panorama or b) give me a wider view with a specific lens by horizontal shifts with the back in portrait. Final format is 54x72 vs native 40x54.

I concur with what's already been said about the relative advantages / disadvantages of each approach. For me, the important advantages of flat stitching are:

• It is very quick. I can complete the image captures in just a few seconds.
• Long, flat horizons (like the ocean) don't get distorted at all by stitching software. I just use Photomerge in Photoshop and select "Reposition." That doesn't warp images.
• As Steve mentioned, you get to accurately see and assess the boarders of the finished image in-camera. There is no guessing whether everything will be covered correctly after the software does its thing.

The disadvantages, primarily image quality at the edges of the image circle used, don't apply to me because I specifically select lenses that can handle what I need. To me, this image circle evaluation is my most-valued criteria when making lens choices.

Dave

Very well put.

There is possibly another related issue, involving the difference between elevations and perspectival views. While all photo images involve perspective, the assembly of a set of shifted images has all the images using one point of view - akin to an architectural elevation - a straight on projection.

Assembled of rotated images (taken by rotating the camera) combine views with different perspectival orientation to the subject. Software is clever, but only to a point. The worst case of this is the iPhone panoramic view, which have much distortion (which we tend to accept happily) - but does not yield a true elevation.

To a lesser degree, more carefully constructed rotated images have this same issue, although depending on the software settings selection (PS gives a few different ones) it can be more or less pronounced. But it is different than the "collage" used to assemble shifted images, which involves no perspective corrections in the software.

The use of shifting gives (IMHO) a more relaxed view, especially for a straight on shot of a building. It basically replicates the taking of a single image with a wide angle lens, however stitching allows one to use longer lenses for a different "feel", without the wide-angle look.

 

[vjbelle]

I have compared the two methods many times. When I was shooting with my STC I made use of rise/back fall with the camera movements and then panned to make 16X10 images. I was never concerned about nodal location and never had any issues. When I moved on to a bellows camera I could take advantage of compound movements and used those exclusively. The end results are very similar mainly because I limit my horizontal shifting to 10mm LR and similar pan.

Since my movements aren't severe I am always in the sweet spot of my lenses which mimic Dave's. My advice is not to be afraid of using either method and if you have any concerns then test first.

Victor B.

 

[vjbelle]

Aside from the many excellent posts I would like to add that once compound movements are used it is critical to know the limitations of the lens. Soft corners and edges coupled with vignetting can be worse than dealing with the disadvantages of panning. I've always felt that it was like a 'Pick your poison' proposition which is why I don't necessarily stick with one method.

Victor B.

 

[Paul2660]

From someone who has stitched for years, since 2000 or so, when I first moved to digital (stitching for more output resolution) I find that I use the twist method more often than not. I agree that flat shifting non multi row, will align better and faster. But on a rm3di, max shift is 15mm, unless you rotate the entire camera (and use rise for shift and it's not easy to do at least for me). 15mm will not get what I want for a pano most of the time. I also have pretty much stopped for now using the IQ4/tech solution due to the massive weight and requirement for a tripod all the time (again for me). With the Fuji GFX or X2D, it's very easy to hand hold a massive twisted stitch single row. I don't attempt multi row due all the issues within the stitching, (wind, light changes DOF etc.).

Modern stitching software will usually stitch together a good result, that while not perfect was quick, easy to shoot and stitch later on. I may to manually crop into the shot (For my landscape panos, not a problem or manually warp the image with in photoshop). You can get by with quite a bit of manual warp before you start to degrade the image with 100MP or more.

I have plenty of 6 to 7 vertical image stitches (all into one large horizontal) that work out fine.

As for the nodal point, for years I attempted to also set it for all my lenses and carry a rail to allow for it. I find for my work, it doesn't matter. If I was working in architecture (indoors or out) where such alignment issues matter more I would.

Also, working with what tech glass I have, the 32mm Rodie I have is good to about 12mm of shift, after that the amount of elongation and flattening of subject matter pretty much ruins the shot even outdoors. The 60mm Schneider can easily go to 20mm of shift (camera can't) and the 90mm Rodie can go to maybe 25mm max (camera can't).

If I wanted a multi level solution, I would go back to the tech camera. For me even with a cube, trying to get a 9 shot (3 low, 3 mid 3 high all horizontal) is hard to due without a tech camera.

Paul

 

[scho]

I am one of the few people that like the 1:1 format so I use a Fotodiox Vertex adapter for Mamiya 645 lenses with the Sigma fpL camera. Four rotated shots (60 MP each) after stitching in LR Photomerge yields a 150 MP 1:1 image. I usually shoot at f/11 or f/16 for DOF. Quick and easy

 

[Geoff]

Seems like rotation is workable for many folks, and that's a lot easier. Staying within the lens limits, not overdoing it, perhaps shooting landscape rather than architecture helps too. Probably software has gotten much better at too. My interest is in careful aligments of vertical and horizontal lines, and thus the preference for shifting. Glad others have found other workable answers.

 

[dougpeterson]

Generally when I do flat stitching behind the image circle I feel like I'm "photographing". When I'm doing pan-and-stitch I feel like I'm doing "computational imaging". These are not perfect words for what I'm trying to describe... there's a feeling of intentionality, relaxation, and connection to the equipment and subject when I'm flat stitching while rotation stitching leaves me feeling detached, clinical, and generally unfulfilled. It's crafting art vs performing a tedious technical chore.

Personally I prefer two-shot flat stitching – whether long side for a panoramic or short side for a square – or maximum three-shot. As soon as I do 2x2 or higher flat stitching it starts to feel like a chore again and more like rotational stitching.

This is 100% subjective and entirely specific to how they make me feel. I would not begrudge a soul for feeling differently. There are many roads that lead to Rome.

 

[Steve Hendrix]

Seems like rotation is workable for many folks, and that's a lot easier. Staying within the lens limits, not overdoing it, perhaps shooting landscape rather than architecture helps too. Probably software has gotten much better at too. My interest is in careful aligments of vertical and horizontal lines, and thus the preference for shifting. Glad others have found other workable answers.

Well this is a critically relevant point, Geoff, and points to the phrase "doing a pano". Generally, rotation speaks to enlarging the "scenic-ness" or left and right dimensions of a scene to create a wider view. It does not address correcting vertical lines, it does not address precise framing, it does not address compositional decision-making, particularly in terms of the amount of foreground and background, as well as subject to frame ratios.

To me, rotation is a beneficial, user friendly tactic (from the standpoint of field use and equipment acquisition), but it falls short of the utility a flat stitching camera provides. Which means there are solid choices for all, but it's important to understand and appropriately assess the pros/cons for each. Hence, this thread represents a good conversation for anyone interested in this type of photography.


Steve Hendrix/CI

 

[Paul2660]

What it boils down to for me, is how much time in post will be required to get an image that I can use. Net, flat field with a tech camera takes a lot of time, and depending conditions you may not even be able to finish. In the old days of manual stitching, back to a Zork adapter with Mamiya glass on Canon, I did all flat field. The stitches were OK, but the Zork had issues with max shifts. Manual combination, took even more time. Once Photoshop started with their Panomerge options, way back to CS3, things picked up.

The software has just gotten better over time, and Ptgui can pretty much stitch anything, it may not be perfect, but much of that can be fixed in post.

Personally I tend to prefer a 3:1 over 3:2 ratio, when shooting as I can always crop into the image and get any other ratio I need. Maybe I am just getting old (for sure) and lazy, but it the old issue of time taken (capture and post work) for reward. I also fully understand the needs for straight lines and verticals to line up with certain subject matter and realize that my methods are not the best for those, however it can work at times.





First example is a 8 part vertical stitch, all rotational on a tripod, with a D850 and 24-70 lens. Main created for resolution, recently printed at 36" x 80". The 2nd image was taken with a Leica SL2 and 24-90 lens, hand held in 4 vertical stitches.

Could both have been taken in one frame or max two? Sure, but not hold up to the resolution required for a large print.

My point is that end result can be gotten with much less effort and equipment. It's all a matter of what you prefer.

Paul

 

[dchew]

What it boils down to for me, is how much time in post will be required to get an image that I can use. Net, flat field with a tech camera takes a lot of time, and depending conditions you may not even be able to finish. In the old days of manual stitching, back to a Zork adapter with Mamiya glass on Canon, I did all flat field. The stitches were OK, but the Zork had issues with max shifts. Manual combination, took even more time. Once Photoshop started with their Panomerge options, way back to CS3, things picked up.

The software has just gotten better over time, and Ptgui can pretty much stitch anything, it may not be perfect, but much of that can be fixed in post.

Personally I tend to prefer a 3:1 over 3:2 ratio, when shooting as I can always crop into the image and get any other ratio I need. Maybe I am just getting old (for sure) and lazy, but it the old issue of time taken (capture and post work) for reward. I also fully understand the needs for straight lines and verticals to line up with certain subject matter and realize that my methods are not the best for those, however it can work at times.

View attachment 217976

View attachment 217977

First example is a 8 part vertical stitch, all rotational on a tripod, with a D850 and 24-70 lens. Main created for resolution, recently printed at 36" x 80". The 2nd image was taken with a Leica SL2 and 24-90 lens, hand held in 4 vertical stitches.

Could both have been taken in one frame or max two? Sure, but not hold up to the resolution required for a large print.

My point is that end result can be gotten with much less effort and equipment. It's all a matter of what you prefer.

Paul

Hi Paul,
I could have written the exact same post, but would replace “flat” with “rotational” and vice versa.
Different strokes for different folks.
Dave

 

[darr]

Well this is a critically relevant point, Geoff, and points to the phrase "doing a pano". Generally, rotation speaks to enlarging the "scenic-ness" or left and right dimensions of a scene to create a wider view. It does not address correcting vertical lines, it does not address precise framing, it does not address compositional decision-making, particular in terms of the amount of foreground and background, as well as subject to frame ratios.

This is why some of us choose to invest in high-end tech cameras with lens or sensor movements—they provide, in my opinion, the best tools for panorama creation. For me, another crucial element in crafting panoramas is lens selection. While wide-angle lenses are indispensable in my toolkit, they often lack the perspective I seek in my panoramas. I tend to favor normal to telephoto lenses because they allow me to achieve a more intimate and focused view, which cropping alone cannot always deliver.

While rotating from a tripod head is a common approach, it's not always feasible for the types of images I aim to create, further fueling my appreciation for tech cameras.

Here's a recent three-image stitched panorama I shared in the medium format image thread. The process was effortless in Lightroom, with no stitching cleanup required since the slices aligned perfectly. For me, the longer lens delivers the intimate perspective I desire to fully convey my connection to the scene. This particular image was created using an STC and the Schneider 72mm f/5.6 L lens.

 

[stevev]

Beautiful image darr.

 

[darr]

Beautiful image darr.

Thank you, Steve. It was a beautiful place to be with a camera.
Best to you!