Terrible stitching errors

[Ben Coffman]

 

Hi,

 

I took this panorama a few days ago and PTGui Pro (version 10.0.15) can’t seem to stitch it at all. Photos were taken with a Canon 6D and Rokinon 24mm lens on a tripod and Nodal Ninja 4 pano head. Past pano projects with the same setup (under similar conditions) have yielded perfectly fine panos, but this time the stitching is a mess. I thought maybe I had accidentally shifted my lens from the no parallax point, but Lightroom, Photoshop, and Microsoft ICE have all yielded panoramas with few or no stitching errors from these files, but PTGui isn’t even close. Any idea what I’m doing wrong in PTGui or what setting I’ve gotten wrong?

 

I’ve uploaded small jpegs of the project here: https://www.dropbox.com/sh/iq0c8uc3n07d07a/AAAXc842C_gB8XlyM3xgsDBTa?dl=0

 

Let me know if you’d prefer larger file sizes.

 

Thanks in advance for your time!

 

-Ben

[Erik Krause]

Am 28.09.2016 um 19:23 schrieb Ben Coffman:
> Any idea what I’m doing wrong in PTGui or what setting I’ve gotten
> wrong?

PTGui does a decent job aligning the majority of the images
automatically. It has a hard time to align the rest because of lack of
features or lack of contrast. If you (hopefully) shot raw, try to
convert to a set of images with higher contrast and boosted shadows.
PTGui will probably align more images. Once you have a successful
alignment you can replace the images with more pleasing ones.

Since you know the shooting pattern you can roughly align the images
which PTGui has problems with. Then manually set control points between
those images and the adjacent ones. Set 2 point pairs for the time
beeing. After an optimizer run you can use the Create Control Points
Here feature to create more.

You shot much too much images. 10 in a row would have done perfectly
giving you 32% overlap. Even 8 would be possible (15% overlap), if you
have good lens parameters. With 15 in a row you have 55% overlap.

There is a very good article about night sky photography by the master
of night panoramas, Aaron Priest:
http://galleries.aaronpriestphoto.com/Articles/Night-Sky-Photography

--
Erik Krause
http://www.erik-krause.de

[Ben Coffman]

Thanks for the feedback; I'll definitely look into the Create Control Points Here feature.

In my defense, normally PTGui does fine with this workflow. (I shoot with more overlap for these night shots, and in the past it hasn't been an issue.) Obviously the black volcanic rock is causing PTGui problems, but why doesn't it cause problems for Lightroom? Or Photoshop? Or Microsoft ICE? All of those programs stitched the pano on the first try.

Also, I guess I didn't see anything about panoramas in Priest's blog post that you posted. Did I miss something?

[Erik Krause]

Am 28.09.2016 um 22:40 schrieb Ben Coffman:
> In my defense, normally PTGui does fine with this workflow. (I shoot with
> more overlap for these night shots, and in the past it hasn't been an
> issue.)

It makes shooting times longer than needed and since stars move it might
cause alignment problems.

> Obviously the black volcanic rock is causing PTGui problems, but

The rock was aligned perfectly in my try, except one image where it was
grey instead of black. The not aligned images where: 0-8, 12-15, 17-22,
28 and 33. With about 20 minutes of work I got this:
https://www.sendspace.com/file/rsq6ny
It's preliminary only, some images don't have control points yet and
some have too few, but it should be a starting point.

> why doesn't it cause problems for Lightroom? Or Photoshop? Or Microsoft
> ICE? All of those programs stitched the pano on the first try.

Joost must answer this. PTGui generally assumes more degrees of freedom
for the placement of images than the other programs, which unfortunately
means higher demands for the images. But this is just a wild guess...

> Also, I guess I didn't see anything about panoramas in Priest's blog post
> that you posted. Did I miss something?

He mentions panoramas a few times but you are rigth: It's more general
about night sky photography. However, if you master single night sky
shots and know about panoramas in general there should be little to add.

[John Houghton]

On Wednesday, September 28, 2016 at 9:40:53 PM UTC+1, Ben Coffman wrote:

In my defense, normally PTGui does fine with this workflow. (I shoot with more overlap for these night shots, and in the past it hasn't been an issue.) Obviously the black volcanic rock is causing PTGui problems, but why doesn't it cause problems for Lightroom? Or Photoshop? Or Microsoft ICE? All of those programs stitched the pano on the first try

Since the sky is spinning around, you cannot simultaneously get good alignment of the sky and of the land features.  The sky seems to be the most important feature, and that can be aligned extremely well at the expense of  the land.  The alignment of the latter is less important given that it is so dark.  Any misalignments of the horizon are much more obvious but can be dealt with masking and minor edits in Photoshop if necessary.

John 

 

[Ben Coffman]

Thanks, Erik, I appreciate your effort. I couldn't see the file in sendspace; do I really need to add a credit card to create an account?

Just for reference, this is my result after my initial alignment: https://www.dropbox.com/sh/211f8gkyq3zlc6j/AAD-N2SEt3muzpOqBJS923FFa?dl=0

The difference between that and the Lightroom-assembled pano is startling. Here's the Lightroom one for reference: https://www.dropbox.com/sh/itghq2b2z973feh/AADvUyu_FUmdLRtAY3O4x47ua?dl=0

Those were exactly the same files, by the way. On mine, 0-8, 12-15, 17-22, 32-34 didn't receive any control points.

If I can bother you further, what were your next steps after "Align images"? Did you go to the Panorama Editor and "Edit Individual Images" to put them in roughly the correct place? Thanks again for your help!

[Ben Coffman]

John, I beg to differ. You can simultaneously shoot the sky and land. I've created dozens of night-sky panoramas, many of them 360-degree, all shot with the same equipment and the same in-field workflow. PTGui usually does a fantastic job stitching them, with no extra masking or cloning needed in Photoshop. Here's an example of a photo I took this summer that PTGui stitched perfectly: https://www.flickr.com/photos/ben_coffman/27657435793/in/dateposted-public/

[Erik Krause]

Am 28.09.2016 um 23:34 schrieb Ben Coffman:
> Thanks, Erik, I appreciate your effort. I couldn't see the file in
> sendspace; do I really need to add a credit card to create an account?

It shouldn't be necessary to create an account.

> Just for reference, this is my result after my initial alignment:
> https://www.dropbox.com/sh/211f8gkyq3zlc6j/AAD-N2SEt3muzpOqBJS923FFa?dl=0

Did you set the focal length?

> The difference between that and the Lightroom-assembled pano is startling.
> Here's the Lightroom one for reference:
> https://www.dropbox.com/sh/itghq2b2z973feh/AADvUyu_FUmdLRtAY3O4x47ua?dl=0

Startling, indeed.

> If I can bother you further, what were your next steps after "Align
> images"? Did you go to the Panorama Editor and "Edit Individual Images" to
> put them in roughly the correct place? Thanks again for your help!

Yes. I deduced from the gaps what numbers the missing images must have
and moved them to the appropriate place. Then on Control Points tab I
set two point pairs between adjacent images of the top row. In the next
row I mainly used the horizon of images below.

[Ben Coffman]

Thanks again, Erik. I've tried both to set the focal length at 24mm in the camera/lens data exif pop up and I've also cancelled out of that pop up and let PTGui figure out a focal length of "25.74 mm" for me. Neither option seems to work well (for me, anyway).

I was hoping there would be a way to put this together with a little less manual work, but I think I'll have to take your advice and start editing the individual images and start adding control points to all of the images that didn't receive them. I'm just really surprised that the other software I tried can get that part right in seconds. PTGui has been my go-to software for these types of panoramas, and I haven't experienced it failing so badly before.

[John Houghton]

Ben, I didn't say you couldn't get a good stitch.  All I was pointing out was that you could not get perfect alignment of the images because of the motion of the stars.  If you align the sky perfectly, the misalignment of the land features is up to 14 pixels on these small images, which equates to 76 pixels on the full size images - far from perfect.  There will inevitably be some stitching errors but you have some control over where these will be according to the positioning of control points and seams.  Misalignments in the sky are unlikely to be noticeable unless there are obvious duplicates of recognisable features like star clusters.

John

[Ben Coffman]

Thanks, John, I misunderstood. I'm assuming that most of my night-sky panoramas have had slight imperfections hidden somewhere in the foreground, which is why they didn't require any masking or cloning in PS.

[John Houghton]

On Thursday, September 29, 2016 at 4:50:21 PM UTC+1, Ben Coffman wrote:

Thanks, John, I misunderstood. I'm assuming that most of my night-sky panoramas have had slight imperfections hidden somewhere in the foreground, which is why they didn't require any masking or cloning in PS.

Ben, FWIW, the result I got can be found here:  https://www.sendspace.com/file/c9kurs . 

John

[Ben Coffman]

Thanks again for your time on this, John. Was there anything usual or special you did? Any Optimizer settings I should know about?

[John Houghton]

On Friday, September 30, 2016 at 12:35:34 AM UTC+1, Ben Coffman wrote:

Thanks again for your time on this, John. Was there anything usual or special you did? Any Optimizer settings I should know about? 

I output two separate stitches for the sky and land with an optimum optimization for each, and merged these in Photoshop with minimal cloning at the line of the horizon to correct a few glitches. The two separate optimizations gave average control point distances of less than 1.  Incidentally, the images in the bottom row appeared to be shot at a different focus setting, since the features at the horizon are clearly out of focus.  The different focus setting results in a slightly narrower field of view, so in theory individual lens parameters should be used for that row.  The fov difference is 0.3 degrees, which in practice affects the stitching only very slightly. My stitch did not take account of this.

John