Nodal point help with panorama

[Ben Rubinstein]

Hi,

I've been trying my first proper nodal point rotation pano's. With my fisheye the pano's are perfect. I worked out where the nodal point (entrance pupil, whatever) was using this tutorial: http://www.johnhpanos.com/epcalib.htm for my 16-35L @ 16mm. I used both methods and came up with the same answer. So far so good.

Problem is that with interiors Autopano refuses to give me a good stitch, even with manual control points there are some 4-5 severe alignment issues with the stitch and after hours of playing I still can't get any better. It's as bad as when I was doing non nodal point rotations.

Can anyone advise?

Here are the files www.studio-beni.net/pano.zip

p.s. PTGui does it fine IF I add control points but I'm stumpted as to why it should be necessary.

 

[Cem_Usakligil]

Hi Ben,

I took a quick look at your files. As far as I can see, you have quite a few challenges which are not related to the entrance pupil. The rotation of the lens around the entrance pupil does indeed prevent parallax problems but it does not help with the deformation of the image between frames due to the combination of rotation/wide-angle/close quarters. I know from experience that using wide angle rectilinear lenses in such close quarters is a huge challenge (i.e. next to impossible). The more detail one has on the foreground (which you have in abundance), the more difficult it is to create a good stitch and blending. Also, you should have at least 50-60% overlap between the frames, preferably even more overlap. From what I can see in these pictures, the overlap is around 30% which is by far not enough. Lastly, I would have gone for a bit longer focal length and shoot more frames, but it would make stitching somewhat easier for this setup. Or you can just use the fisheye lens which should work just fine.

Cheers,

 

[Valentin Arfire]

try this

As my experience with Autopanopro is very limited, I rely upon PTGUI (or hugin) and I get predictable results of it.

I usually try to get a precision after optimisation of less than 2.8 (is pixel difference between near pictures) which I got with a few progressive optimisations.

I will upload the panorama as soon as I will finish it but unfortunately for better results you need multi-row for this specific conditions - 16 mm on your camera : I suppose the low row at -30 degrees, one at 0 next one at 45 and one shot straight up at 90. When taking only one row to obtain a cylindrical panorama it is expected to go perfect with absolutely no stitching errors. Just keep in mind to maintain somewhere 10-23% overlap.

After setting up some nice large halls with plenty of details you'll get the best conditions to set up the best npp, focus and stop. BTW the npp is slightly different when taking 6 or 8 pictures

here is the link to my ptgui project
check it out - can save the lenses parameters to use later - of course wil be much better after optimisation of multi-row panoramas
http://valentin.europhoto.ro/transfer/BenPanorama.pts

 

[Asher Kelman]

I must be missing something here! Are you having an issue with your tour or the nodal point? For flat stitching, where's the problem?

Of course, one can move the center of the pano.

However, it must be that I am overlooking something obvious to everyone else!

Asher

 

[Doug Kerr]

Hi, Asher,

I must be missing something here! Are you having an issue with your tour or the nodal point?

Perhaps you meant to refer to the noodle point.

It is easy to locate by inspection:

Douglas A. Kerr: The noodle point​

Best regards,

Doug

 

[Ben Rubinstein]

Hi Asher,

Try making that pano and look through the completed one at around 50%, there are some 4-5 serious alignment issues.

Cem, I have to try again with more overlap, I had been convinced that the errors were due to an incorrect nodal point, perhaps I should try it with multilayer and 20mm?

What I don't need is a gigapixel stitch, the 20 or so megapixels I get from stitching with the fisheye is what I'm going to be using most of the time anyway so I hardly need more than than the 60 megapixels I get with that 16mm neverming the silly amount of resolution possible with a 50mm lens!

Valentin, I'll have a look when I get back to my computer, I would prefer not to have to purchase PTGui though (I have it on trial mode), autopano should be able to do this methinks, I just have to work out how.

Doug, glad it's on a cheap sigma!

Thanks for the help guys, I'm sure I'll crack it, just needs perserverance, keep the help coming!

 

[Ben Rubinstein]

OK, tried it with multi level, both PTGui and Autopano made the same dogs dinner of the stitching.

Valentin, your version is perfect, I need to work out with you how you did it, perhaps we can arrange some time on skype or something and you can talk me through it?

 

[Bart_van_der_Wolf]

OK, tried it with multi level, both PTGui and Autopano made the same dogs dinner of the stitching.

Valentin, your version is perfect, I need to work out with you how you did it, perhaps we can arrange some time on skype or something and you can talk me through it?

Hi Ben,

I gave your files a try. You have your no-parallax-point (NPP) of rotation set up pretty good, so that's not the issue. While it is possible to make a pretty good stitch (I use PTAssembler), a difficulty for an automatic stitcher lies in finding accurate control points. The rectilinear lens projection with extreme wide-angle lenses will show some items in the overlap region quite differently (due to geometric distortion by projection on a flat plane). You can inprove that situation by making the overlap a bit larger, say 10 images (instead of 8) to go 360 degrees.

To make things much easier, also to prevent having to reshoot a scene, I usually take 50% overlaps in these extreme scenarios. That will create 3 overlaps for each part of a scene, with plenty of control points at each overlap with more similar distortions, so it should be possible to get 'perfect' alignment. It often also helps in scenes with moving objects to automatically eliminate ghosts.

Another trick that might help in perfect blending where the left and right side meet in a 360 degree equirectangular view, copy one of the edge images (e.g. the right edge) so you get a duplicate. Now let the software create control points for that image with the left most image and the right most (identical) image.

Cheers,
Bart

 

[Ben Rubinstein]

Tried the same stitch using 16 frames, no luck in either PTgui or AP. There has to be a way to get a foolproof stitch without having to add loads of control points doesn't there or am I dreaming?

BTW Bart, many thanks for the reassurance about the nodal point positioning, thought I'd go crazy!

 

[Doug Kerr]

Hi, Ben,

BTW Bart, many thanks for the reassurance about the nodal point positioning, thought I'd go crazy!

I have asked Carla to prepare me some soothing tea to allow me to keep my composure amidst all this talk of nodal points.

She says that perhaps some chicken nodal soup would do as well.

Best regards,

Doug

 

[Ben Rubinstein]

Hi, Ben,


I have asked Carla to prepare me some soothing tea to allow me to keep my composure amidst all this talk of nodal points.

She says that perhaps some chicken nodal soup would do as well.

Best regards,

Doug

You got to have pics of needles on your lens for all this talk of stitching..

 

[Doug Kerr]

Hi, Ben,

You got to have pics of needles on your lens for all this talk of stitching..

Good point! We'll look into that.

Best regards,

Doug

 

[Asher Kelman]

To make things much easier, also to prevent having to reshoot a scene, I usually take 50% overlaps in these extreme scenarios. That will create 3 overlaps for each part of a scene, with plenty of control points at each overlap with more similar distortions, so it should be possible to get 'perfect' alignment. It often also helps in scenes with moving objects to automatically eliminate ghosts.

So what overlap would you have for the 24 mm T/S using it only as a fixed lens. Does one have to have more overlap for closer objects.

Another trick that might help in perfect blending where the left and right side meet in a 360 degree equirectangular view, copy one of the edge images (e.g. the right edge) so you get a duplicate. Now let the software create control points for that image with the left most image and the right most (identical) image.

Does one simply have two copies or do you add it after APGiga has made ts first attempt.

Asher

 

[Asher Kelman]

OK, tried it with multi level, both PTGui and Autopano made the same dogs dinner of the stitching.

Valentin, your version is perfect, I need to work out with you how you did it, perhaps we can arrange some time on skype or something and you can talk me through it?

Ben and Valentin,

Let's all benefit from this! So post any ideas and solutions here!

Asher

 

[Ben Rubinstein]

I spent an hour or so on skype with Valentin and the basic solution is to spend about an hour or so optimizing each pano in PTGUI using a 20 step workflow which has frightened the pants off me - or just use the fisheye

This stuff ain't easy! Huge amounts of respect for the experts like Valentin, not only for his knowledge but his willingness to pass it on. It isn't however something that can be summed up in a forum post methinks, more like the subject for a 3 day workshop!

 

[Valentin Arfire]

thinking back to the time I started panoramas with the kit lenses in canon 350 D back in 2006, I believe not too much has changes except I needed for my first panorama of 38 images and from the very beginning to the end it took me more or less of 3 weeks

...hugin and a lot of optimisations

I think this is http://www.360cities.net/image/invingatorul-monument-timisoara#335.10,0.80,70.0

 

[Bart_van_der_Wolf]

So what overlap would you have for the 24 mm T/S using it only as a fixed lens. Does one have to have more overlap for closer objects.

Hi Asher,

Hard to pin an exact number on it. It has to do with the subject matter as well. When there are dots in an image, they more or less remain looking like dots even in the stretched corners. When there are squares, they may end up looking like elongated diamonds, that's when the automatic control points may not be set correctly, or not found at all. Mind you, with some manual intervention it is possible to force Ben's example into submission, but his goal is to do it automatically.

What is sure, is that with 50% overlap you will solve many potential issues, at the expense of a few more tiles and a bit longer shooting time.

Does one simply have two copies or do you add it after APGiga has made ts first attempt.

Adding a duplicate image at one opposite end will allow a blending algorithm that doesn't take 360 degree panos into account to seamlessly blend smooth gradients. So one can take a duplicate of the left most tile, and add it to the right most tile, thus ensuring that the blend is done well where the left and right tiles meet to complete the circle. It can be added from the getgo, but not all blending algorithms need it.

Cheers,
Bart

 

[Doug Kerr]

It is thrilling to me to see the degree of sophistication in technique and artistry of our members, and their stunning results, in this specialized regime.

The tools in their own right are amazing as well.

Kudos to all who labor in these fields, both photographers/artists and tool developers.

Best regards,

Doug

 

[Ben Rubinstein]

OK,

Valentin built me what he described as a 'basic' lens profile for the 16-35L II @ 16mm for use in PTGui. He told me he'd build me a better one once I give him a mutirow pano of the inside of a big building to work with. He also gave me instructions how to build my own profiles.

His 'basic' profile together with his step by step instructions has COMPLETELY solved my problems and I am now getting 100% perfect pano's @ 16mm even with mutirow stitches with this which is about the hardest type of pano to do.

Valentin should be leading a workshop on Panoramic Photography, he is a genius!

 

[Ben Rubinstein]

Can anyone tell me in PTGUI, once I've created an HDR pano (3 brackets of each frame) how I can then tell it to save each 'bracket' as a seperate file so that I have 3 files, one dark, one normal and one light to then input into my HDR program for tone mapping? I can do it as 3 layers in one PSD file but that's a pain to seperate but can't work out how to output 3 seperate files.

 

[Valentin Arfire]

in "create panorama" tab you'll find blend planes, individual layers and exposure fused; when stitching the panorama you'll ontain whatever you want

 

[Valentin Arfire]

I'm not genious. And I hope I haven't reached my roads end.

having a little more experience offered me the chance to do some help I have some recomendations - which I did - the result is yours Ben and of course you're not reaching the end of the road but only its beginning.

 

[Ben Rubinstein]

Here is the same scene though with fresh frames, processed through PTGui with Valentin's profile and HDR'd with 3 versions of the panorama through SNS-HDR. Nothing else done to it.

It took 3 days but I'm a happy bunny...

 

[Ben Rubinstein]

Note though, boy but SNS-HDR is slow and noisy. Took 2.5 hours to build the preview for the three 54 megapixel brackets (Dual core 2.2ghz with 8gb RAM) methinks a 64 bit version would be very welcome in the future! The noise even with these iso 400 shots would put me off printing the images without very careful and significant noise reduction, it's worse than native iso 3200 on my 5D's.

The guy who wrote the zero noise software showed that he had used zero noise to do the initial blend of the brackets which kept the noise right down and then did the tonal mapping using SNS-HDR by getting zero noise to output 3 versions of the same file at 2 stop intervals. Has anyone got any idea how to do this last part?

Perhaps I should not sharpen prior to stitching, it does help the stitching process as it gives the software better information for joins but it is no doubt contributing to the noise problem.

Just another addendum, one of the Autopano team posted on a thread I had put on their forum on the subject of the problems I was having that APP has recognised problems with the distortions from ultra wide rectilinear lenses but they are working on the problem and looking at a comrehensive solution by v2.6 with the inital results looking very promising!

 

[Asher Kelman]

Here is the same scene though with fresh frames, processed through PTGui with Valentin's profile and HDR'd with 3 versions of the panorama through SNS-HDR. Nothing else done to it.

It took 3 days but I'm a happy bunny...

This is wonderful. Now tell us how to do it, step by step!

I'm jealous!

Asher

 

[Ben Rubinstein]

Valentin's PTGui tutorial is some 20 steps and I'll leave it to him to decide whether to post it up, it's his work to take pride in, not mine, however..

a) With scenes like this it's essential to have a nodal slide. Period.

b) I metered for a good 'normal' exposure then bracketed by 2 stops either way.

c) Processed the RAW's so that the WB, etc was the same and kept the contrast/curve neutral for maximum information. I also capture sharpened though I'm wondering whether it might have been a mistake. I kept the Adobe Lens Corrections firmly off. I saved as jpg but in the future will use Adobe '98, and 16 bit Tiff's from the start for maximum information and to make sure my computer really crawls!

d) Processed the files through PTGui using Valentin's profile and over 50 rounds of optimization (as instructed). Output as 3 seperate files (once I work out how) or as a PSD with 3 layers.

e) Plug in to HDR software of choice.

f) Post up here and pretend I am in any way an authority on the subject after doing my first ever HDR panorama using PTGui.

 

[Ben Rubinstein]

Here is the crucible of a pano. A tiny (and messy, we're mid spring cleaning) balcony just 6X9 feet (2X3m) in size with near objects, a high set of railings and distant subjects. We also have Middle East sun, flare, etc.

Autopano got it about 90% right eventhough I was using the fisheye. PTgui using Valentin's instructions got it 100% perfect when I was using the fisheye and then when I used the 16-35L (with his profile), I hadn't expected the 16-35L to stitch so perfectly with what is a very difficult subject!

As you can see SNS-HDR has done an admirable job with the final output, so natural looking! I didn't sharpen beforehand this time and the noise is far better controlled.

This is from the 8mm. I screwed up a little bit and forgot Valentin's advice to shoot at -15 degrees then a zenith shot so there is a tiny bit of error at the zenith and nadir when making into a virtual tour but hey, this was just a trial anyway. I would love to know how to reduce the aliasing/moire when rotating the tour though, drives me crazy...

http://www.studio-beni.net/s4.html

 

[Bart_van_der_Wolf]

Note though, boy but SNS-HDR is slow and noisy. Took 2.5 hours to build the preview for the three 54 megapixel brackets (Dual core 2.2ghz with 8gb RAM) methinks a 64 bit version would be very welcome in the future!

Hi Ben,

SNS-HDR is both 32-bit and 64-bit, it automatically installs the 32-bit or 64-bit version from a single installer package depending on one's OS. It also uses all cores in a multiprocessor CPU, and it utilizes the GPUs of the graphics card if possible, so a beefy Graphics card also helps.

It can only enhance noise if it's there to begin with. Therefore I tend to use it on ISO 100 brackets, or denoised images. The results look so natural because a lot of calculations need to be done to achieve that. You probably understand why I suggested to do the tonemapping per tile, then stitch/blend the tiles into a panorama. It also allows to vary the color balance per tile. With a little preparation you can process the tiles in a batch with SNS-HDR Professional version.

The noise even with these iso 400 shots would put me off printing the images without very careful and significant noise reduction, it's worse than native iso 3200 on my 5D's.

Why ISO 400? Also, look at the settings you use in SNS-HDR, it's easy to get carried away with midtone contrast and highlight contrast.

The guy who wrote the zero noise software showed that he had used zero noise to do the initial blend of the brackets which kept the noise right down and then did the tonal mapping using SNS-HDR by getting zero noise to output 3 versions of the same file at 2 stop intervals. Has anyone got any idea how to do this last part?

Guillermo Luijck from Spain has written the Zeronoise software, and at the bottom of that page he describes the Auto AEB output option (click on the UK flag for a Google translation).

Perhaps I should not sharpen prior to stitching, it does help the stitching process as it gives the software better information for joins but it is no doubt contributing to the noise problem.

Possibly. Just review the whole workflow, how to get a lower noise basis for postprocessing. Remember that having many photons will allow to have a high Signal to Noise Ratio (SNR), meaning one needs to collect lots of photons for low noise images. Bumping ISO will do the opposite, it will use fewer photons. There is also nothing wrong with some careful noisereduction before tonemapping if one has to push the tonemapping settings. SNS-HDR also does some noise reduction if selected at the opening stage.

Depending on the stitching software, it doesn't necessarily depend on sharp features to place its automatic control points. Besides, one can always add manual control points, humans are much better at finding corresponding patterns in differently distorted images.

Cheers,
Bart

 

[Ben Rubinstein]

Hi Ben,

SNS-HDR is both 32-bit and 64-bit, it automatically installs the 32-bit or 64-bit version from a single installer package depending on one's OS. It also uses all cores in a multiprocessor CPU, and it utilizes the GPUs of the graphics card if possible, so a beefy Graphics card also helps.

Fair enough, hadn't realised it was 64 bit but it is indeed. My graphics card is ancient so I suppose that could be a bottleneck.

It can only enhance noise if it's there to begin with. Therefore I tend to use it on ISO 100 brackets, or denoised images. The results look so natural because a lot of calculations need to be done to achieve that. You probably understand why I suggested to do the tonemapping per tile, then stitch/blend the tiles into a panorama. It also allows to vary the color balance per tile. With a little preparation you can process the tiles in a batch with SNS-HDR Professional version.

I would prefer waiting rather than trying to match mutiple slightly different frames though I have to have a play with the batch option.

Why ISO 400? Also, look at the settings you use in SNS-HDR, it's easy to get carried away with midtone contrast and highlight contrast.

You're kidding right? The ability to freeze movement with a stopped down lens has often had me shooting pano's at iso 1600, software that cannot cope with iso 400 in 2011 is a joke. I didn't apply any more than the defaults of contrast in the 'natural' preset. Given that HDR should reduce noise rather than enhance it, it is strange that the program creates about triple the amount of noise in the shadow areas than was present originally in a brighter bracket of the same scene which the program had to use. If I wanted to have the look of a heavily over pushed file then I wouldn't bother bracketing in the first place. The ceiling in that image is horrifically noisy where no noise exists in the image with the same brightness. There just shouldn't need to be any noise at all in the far right of the histogram but with the iso 100 image on the balcony I still had to apply noise reduction to the blue sky. On a 5D that should be unimagineable.

Guillermo Luijck from Spain has written the Zeronoise software, and at the bottom of that page he describes the Auto AEB output option (click on the UK flag for a Google translation).

I have to have a play and also see if the program will work with TIFF files or only RAW files so I can send it finished pano's, thanks!

Possibly. Just review the whole workflow, how to get a lower noise basis for postprocessing. Remember that having many photons will allow to have a high Signal to Noise Ratio (SNR), meaning one needs to collect lots of photons for low noise images. Bumping ISO will do the opposite, it will use fewer photons. There is also nothing wrong with some careful noisereduction before tonemapping if one has to push the tonemapping settings. SNS-HDR also does some noise reduction if selected at the opening stage.

I still maintain it shouldn't be necessary, why bracket if pushing the shadows will give me the same amount of noise? I was of course using the noise reduction.

Depending on the stitching software, it doesn't necessarily depend on sharp features to place its automatic control points. Besides, one can always add manual control points, humans are much better at finding corresponding patterns in differently distorted images.

I had been used to APP which does benefit from presharpened images. Now I've been introduced to PTGui I find that it's significantly more powerful and stitched my unsharpened images without a problem where APP created errors.

Cheers,
Bart

Much appreciated Bart! There is so much to learn, it seems that pano photography is still in many ways on the cutting edge of technological advances.

 

[Bart_van_der_Wolf]

You're kidding right? The ability to freeze movement with a stopped down lens has often had me shooting pano's at iso 1600, software that cannot cope with iso 400 in 2011 is a joke.

For panoramas of interiors, I can understand it could be useful in Japan or other earthquake sensitive venues ...

However, and I kid you not, ISO 400 has at least double (ISO 1600 four times) the amount of shot noise because only one quarter (ISO 1600 1/16th) of the number of photons is being recorded in a well exposed frame compared to ISO 100. What's more is that the dynamic range per shot is reduced (and the effect of veiling glare increased), requiring smaller bracketing intervals as a partial remedy.

I do understand the occasional need to freeze subject motion (mostly outdoors), but bracketing will not be as successful for that scenario either.

I didn't apply any more than the defaults of contrast in the 'natural' preset. Given that HDR should reduce noise rather than enhance it, it is strange that the program creates about triple the amount of noise in the shadow areas than was present originally in a brighter bracket of the same scene which the program had to use.

There is a difference between reducing noise by capturing more photons (and subsequent downscaling of the exposure level), and boosting micro-contrast in postprocessing (which is what tonemapping does). Tonemapping will exacerbate all noise just as readily as microdetail, it can't make a very good distinction. We humans are better judges when it comes to noise reduction, and we also can do things for prevention.

Cheers,
Bart