Panoramic Experiment

[fike]

I knew this fact, but I had to prove it to myself.  You know how that is...

I went out to make an equivalent panoramic field of view using lenses at focal length ranges from 10mm to 70mm.  I made seven versions:

• 10mm
• 13mm
• 22mm
• 24mm
• 35mm
• 52mm
• 70mm

Here are the raw panoramic images put together (ignore differences in color and contrast).




Then I cropped the images to the same (similar) field of view.



As expected, the images are exceedingly similar.  I knew it would work, but I had to prove it.  I did learn a few things about pano work that I didn't expect.

1) I have always argued that a moderately wide lens like the 24-70 could be used to make very wide images with panoramic techniques.  What I didn't spend much time considering was the fact that large expanses of sky or water are very difficult to stitch well when you lack control points.  With an ultra-wide lens, you are more likely to have land features in each image segment, allowing you to get control points.
2) Ultra wide angle lenses are a bit better for dramatic skies because they include the wide expanses of sky in its exposure averaging.  When shooting a panoramic in more components, it can sometimes be difficult to find a good average to shoot, thus necessitating HDR.  This is not to say that wide angle lenses get more dynamic range, just that metering for sky and foreground is a bit easier.
3) 10mm is too wide for panoramic images if you intend a pretty 'normal' rectilinear projection to work.

And just for fun (we could easily have calculated this) two 10mm images stitched together require more than forty images stitched together in two (or more) rows of twenty images at 70mm.

[feppe]

1) I have always argued that a moderately wide lens like the 24-70 could be used to make very wide images with panoramic techniques.  What I didn't spend much time considering was the fact that large expanses of sky or water are very difficult to stitch well when you lack control points.  With an ultra-wide lens, you are more likely to have land features in each image segment, allowing you to get control points.

I struggled with skies as well in the beginning. Now what I do is take the pano normally at whatever relatively long focal length, and take the sky separately at a much wider focal length. At least Autopano Pro is able to mix different focal length shots in the same final pano. Obviously you get less resolution in the clouds, but that's ok to me as I don't have to waste hours fixing the seams. The result can be seen on my Hong Kong at night shot - 100mm equivalent lens for buildings, somewhere between 20mm and 30mm equivalent for sky.

[sergio]

I choose my focal length based on output resolution and printed size. The bigger I intend to print, the longer focal length I use, consequently more pixels are used to conform the image. Lens quality and optical design also play a role for me. Some lenses flare more than others and that is a pain when stitching. Very good post and thank you for sharing your discoveries.

[Ben Rubinstein]

From experience FOV relative to subject distance is the only thing that makes any difference at all to perspective in a pano. Take a single shot with the lens that will give you the perspective you're looking for. Then standing in the same spot use a longer lens and take enough frames to fill the same FOV. Hey presto you get the perspective required. I use this a lot when shooting architectural stuff when I want a certain perspective 'look' to my images.

[fike]

Yes, added resolution is generally my motivation for large panoramic stitches.  The smallest sizes I typically print are 13" tall, and more commonly, I print the height of a wide panoramic at 17" or 23".  Based on 240 DPI, that means I need one row to get 17" and closer to two rows (considering overlap) for 23" tall images.  I would say that I often capture too high a resolution, but that has its advantages because I can crop more aggressively and still be left with a large print.  

Part of this exercise has been to help me determine if I want to get a 10-22 lens.  I am not too sure I am sold.  At 24mm, I can shoot three images to every one image at 10mm.  I don't like the distortion on the 10mm, so I would probably pull back to 13mm or 15mm where the ratio to the 24mm lens is more like two images to every one. That isn't really an obstacle to me.  Shooting fewer than 15 images really is pretty quick and easy.  

Someone else brought up flare in the context of shooting panoramics.  Generally the wide angle lenses are more prone to flare than something like my 24-70.  This is another good reason to use a longer lens.  Flare is more controlled, and you can shoot more overlap, providing the opportunity to blend your layers to cover any flare that does occur.  

I really wish there was a modestly-priced and high-quality 15mm fixed lens--something akin to the 50 f/1.4.  I am not sure having a zoom on these ultra wide lenses is much help.  It is really a special purpose lens that you put on for a special affect and take off when you are done.  Smaller and cheaper is my goal for a lens like that.

[walter.sk]

Yes, added resolution is generally my motivation for large panoramic stitches.  The smallest sizes I typically print are 13" tall, and more commonly, I print the height of a wide panoramic at 17" or 23".  Based on 240 DPI, that means I need one row to get 17" and closer to two rows (considering overlap) for 23" tall images.  I would say that I often capture too high a resolution, but that has its advantages because I can crop more aggressively and still be left with a large print.  

Part of this exercise has been to help me determine if I want to get a 10-22 lens.  I am not too sure I am sold.  At 24mm, I can shoot three images to every one image at 10mm.  I don't like the distortion on the 10mm, so I would probably pull back to 13mm or 15mm where the ratio to the 24mm lens is more like two images to every one. That isn't really an obstacle to me.  Shooting fewer than 15 images really is pretty quick and easy.  

Someone else brought up flare in the context of shooting panoramics.  Generally the wide angle lenses are more prone to flare than something like my 24-70.  This is another good reason to use a longer lens.  Flare is more controlled, and you can shoot more overlap, providing the opportunity to blend your layers to cover any flare that does occur.  

I really wish there was a modestly-priced and high-quality 15mm fixed lens--something akin to the 50 f/1.4.  I am not sure having a zoom on these ultra wide lenses is much help.  It is really a special purpose lens that you put on for a special affect and take off when you are done.  Smaller and cheaper is my goal for a lens like that.

Thanks for your interesting experiment.  I have long been interested in panos and have shot typical landscapes with wide lenses, from circular fisheye and rectangular fisheye lenses through about 50mm.  I have also isolated distant scenes such as a railroad bridge at the other end of a lake using longer lenses, upto 400mm.

I have also changed some of my pano techniques with the advent of improvements in stitching programs such as PTGui and PanoTools, as well as Photomerge in CS4.  The more accurate stitching has enabled me to make the following changes:

1)  Instead of manual focus about 1/3 into the scene I now use autofocus on each image in the sequence.  The panos stitch quite well and my plane of focus, say on a row of buildings, is more consistently sharp.

2)  Instead of manual exposure set to save the highlights of the brightest area of the scene, I now use autofocus on each image, which keeps the overall pano within the exposure range I want without having to bring up shadow detail along with noise.

3) I still use manual white balance, set either using the most appropriate preset, or a custom wb, as that still gives the best results.

[fike]

Thanks for your interesting experiment.  I have long been interested in panos and have shot typical landscapes with wide lenses, from circular fisheye and rectangular fisheye lenses through about 50mm.  I have also isolated distant scenes such as a railroad bridge at the other end of a lake using longer lenses, upto 400mm.

I have also changed some of my pano techniques with the advent of improvements in stitching programs such as PTGui and PanoTools, as well as Photomerge in CS4.  The more accurate stitching has enabled me to make the following changes:

1)  Instead of manual focus about 1/3 into the scene I now use autofocus on each image in the sequence.  The panos stitch quite well and my plane of focus, say on a row of buildings, is more consistently sharp.

2)  Instead of manual exposure set to save the highlights of the brightest area of the scene, I now use autofocus on each image, which keeps the overall pano within the exposure range I want without having to bring up shadow detail along with noise.

3) I still use manual white balance, set either using the most appropriate preset, or a custom wb, as that still gives the best results.

Yes, panoramic stitching software has certainly made the process far more forgiving.  In a thread from a few months ago, Pom got me thinking about the necessity of using the nodal slide.  

http://luminous-landscape.com/forum/index....mic+nodal+point

It is only really essential when you are concerned with very close foreground elements or you have many straight lines like you might have in a city setting or indoor setting.

I have also streamlined a few things.  I have modified my camera settings so that the shutter release doesn't control focus. This has enabled me to use the thumb-button (on a canon) to set focus and then stop it from refocusing.  Then I can press and hold the * button to set the exposure and keep it fixed across multiple shots.  By doing this and shooting RAW, I can shoot panoramics quickly while staying in program mode.  

I haven't had good luck with allowing refocusing between images.  I have done some experimenting with focus bracketing between lower and upper rows.  I still haven't mastered it yet.

[RafalA]

Thanks for an interesting experiment and post.

Just getting into stitched panoramics, and I have often wondered about the FL myself and you've saved me the time to do this.

[OldRoy]

Interesting thread. I've spent a lot of time in the last year on stitched panos, concentrating on spherical VRs as this service is something that I can market and get a small return on what is essentially a hobby that I can't really afford. Also I'm interested in old churches and I find that they are a really appropriate application for HDR VR panos, a medium that seems to be a photographic "poor relation". FWIW I have been using D200/ 10.5 FE for VRs. I now also have a D700 and have had the 10.5 lenshood shaved so that I can use it as a circular FE on this body. I'm not convinced about the utility of this BTW.

A few thoughts prompted by this thread.

Doing multi-row panos with long lenses really only seems to make sense once the final print size (not referring to VRs here) means that you're running out of resolution in single row sets. I would have thought that this could be tabulated? At least by someone with a lot of time on their hands! Also, using zooms is problematic if there's any chance of foreground elements - and it frequently makes for a better composition if there are - so it's necessary to map the NPP at a range of focal lengths. Likewise I wish this data were readily available. I have three new lenses and an FX body and balk at the thought of having to do this again. I'm lazy and there must be hundreds of people doing the same thing for the same relatively small range of lenses. Why don't the lens manufacturers supply this information at least for the axial NPP position? The Pano head suppliers like Nodal Ninja do list some users' calculations, but they are mostly confined to FEs for VR use.

One problem I have come up against repeatedly, and it's fairly unavoidable I suppose, is that of curvature in cylindrical panos. This is particularly obtrusive when trying to shoot an architectural exterior at fairly short range where, for example, the best coverage is to shoot 2 or 3 portrait format shots at fairly short focal lengths (24mm for example). It produces a conspicuous curvature of ridgelines. So I've recently acquired a 14-24 2.8 Nikkor which reduces the requirement - and very nice it is too, even if it still requires a lot of perspective correction. I've also experimented with it for multi-row VRs, but I think that it's overkill and hard work for web use. For HDR it can entail as many as 112 shots, given Nikon's idiotic bracketing limitations.

I use PTGui and I'm intrigued by the comment someone made that AutoPano Pro enables one to shoot skies at a different focal length and then stitch this shot. I'm wondering how one gets round the problem of differing resolution at the join - particularly if there is a lot of fine detail such as a treeline - even with a alpha channel mask. My own experience of patching in viewpoint corrected nadir shots (to cover the tripod hole in full 360 x 180 VRs) which also involves effectively different combined resolutions, is that it can sometimes be visible.

Anyway, it's good to see some discussion of one's own enthusiasms!

Roy

[fike]

With the type of work I do, I haven\'t had good results trying to use different focal lengths for the skies.  Perhaps for an urban scene or a distant horizon, it works okay, but I am typically dealing with lots of little branches and leaves that move very small distances and have very small openings between which you can see the sky.  this doesn\'t lend itself to stitching at different focal lengths as it also doesn\'t lend itself to HDR.  My best results with this problem of brighter skies than foreground, is to shoot a lower row for the shadows (typically) then shoot an upper row that overlaps 50% or more and expose for the highlights.  Then I stitch the lower and upper rows separately.  Finally I stitch the rows together and then manually blend the exposures or use a gradient mask to get a different kind of graduated neutral density filter.  The important thing with this effect is that your seams can\'t be blended with transparency because they will never align all those little branches perfectly.  You are better off manually selecting a meandering path that uses the shadows and the landscape to hide your two exposures.  It works better than it sounds.  

Here is my favorite example of this technique

[Luis Argerich]

This is a great thread I think I will come back with something more substantial because I really enjoy panoramas but for now I'd just say thanks!

[OldRoy]

he important thing with this effect is that your seams can\'t be blended with transparency because they will never align all those little branches perfectly.  You are better off manually selecting a meandering path that uses the shadows and the landscape to hide your two exposures.  It works better than it sounds.  

Here is my favorite example of this technique

Thanks for this suggestion Marc, which I'll try at some point. You're right in that I wouldn't have expected it to work very well (or maybe just not in my hands.) I'm assuming that this is a situation where you have exposure variations that wouldn't allow for "artifical HDR" (ie a single exposure setting PPd at 2 different sets of values) ? Or where active D-lighting wouldn't help enough - in the case of a Nikon user anyway.
BTW your example is too small on the site to get much idea what it looks like, except compositionally.
Roy