Synthetic test, with features. How to make it work?  

Share your tips and tricks here or get help with any Autopano Pro / Giga problem!
No bug reports (of any kind) in this forum!
no avatar
pen_or_rhino
New member
 
Topic author
Posts: 9
Likes: 0 post
Liked in: 0 post
Joined: Mon Jul 30, 2018 12:14 pm
Info

Synthetic test, with features. How to make it work?

by pen_or_rhino » Mon Jul 30, 2018 5:06 pm

I cut several images out of this single image:
pano-source.png
pano-source.png (958 Bytes) Viewed 1351 times

Here are the 5 first pieces. I want vertical column because I believe the "theta"-span is more difficult and makes a better test.
pano1.png
pano1.png (295 Bytes) Viewed 1351 times

pano2.png
pano2.png (299 Bytes) Viewed 1351 times

pano3.png
pano3.png (318 Bytes) Viewed 1351 times

pano4.png
pano4.png (294 Bytes) Viewed 1351 times

pano5.png
pano5.png (398 Bytes) Viewed 1351 times


Unlike images from my previous post, in this case panorama is certainly stitchable (there are enough clearly-definable control points).

When I load them in AutoPano (jpeg equivalents, some error for pngs), AutoPano does not recognize them all. Only 4, and places them incorrectly. Independent of that setting, "Force every image..."

How to make it see this panorama?

Thanks.

no avatar
Nanard
Member
 
Posts: 150
Likes: 0 post
Liked in: 3 posts
Joined: Sun Jul 29, 2007 11:00 pm
Location: France
Info

Re: Synthetic test, with features. How to make it work?

by Nanard » Tue Jul 31, 2018 9:23 am

I already noticed that Autopano has more problems with vertical panos.
Last time, what I did was to rotate each image to change the pano from vertical to horizontal.
Sounds odd but it worked better.
Bernard

User avatar
olivierbo
Kolor Team
 
Posts: 2227
Likes: 1 post
Liked in: 132 posts
Joined: Thu Nov 06, 2014 11:44 am
Info

Re: Synthetic test, with features. How to make it work?

by olivierbo » Tue Jul 31, 2018 9:48 am

Hello,

You have a lot of homogenic areas and not enough details to let Autopano detect a correct panorama. Here is how Autopano works:
http://www.kolor.com/wiki-en/action/vie ... _RMS_value
http://www.kolor.com/wiki-en/action/vie ... n_pipeline

This is the same problem with your other post. No details so Autopano cannot find any similarities to stitch the pictures.

Hope this can help.

Kind regards,
Olivier

no avatar
pen_or_rhino
New member
 
Topic author
Posts: 9
Likes: 0 post
Liked in: 0 post
Joined: Mon Jul 30, 2018 12:14 pm
Info

Re: Synthetic test, with features. How to make it work?

by pen_or_rhino » Tue Jul 31, 2018 8:53 pm

Nanard wrote:Sounds odd but it worked better.

No, not odd actually. Horizontal shifting is trivial, shift as is without changing shape, while vertical shifting involves shape distortion - more difficult.

This my 2nd question is actually relevant only if my 1st one can be resolved. Since smooth patterns can't be stitched (in AutoPano) - I'm not interested anymore in this. I apologize for asking this question before I got reply for my first question.

no avatar
Nanard
Member
 
Posts: 150
Likes: 0 post
Liked in: 3 posts
Joined: Sun Jul 29, 2007 11:00 pm
Location: France
Info

Re: Synthetic test, with features. How to make it work?

by Nanard » Tue Jul 31, 2018 10:05 pm

I don't understand what you are saying , nor what you are trying to do.
Are you testing autopano with limit cases, or are you looking for a solution to a real problem ?

no avatar
pen_or_rhino
New member
 
Topic author
Posts: 9
Likes: 0 post
Liked in: 0 post
Joined: Mon Jul 30, 2018 12:14 pm
Info

Re: Synthetic test, with features. How to make it work?

by pen_or_rhino » Wed Aug 01, 2018 7:32 am

Originally I have I real problem that I want to solve.

I am new to panoramic world, so I am trying to make sure basic things make sense to me.

Here's the stitched panorama from above, in Hugin(*), with CPs manually set, with about 30 degrees horizontal FOV:
polar_enlargement2.jpg

Notice that the rectangles get distorted the closer they are to the south pole. This is very basic property, and I was trying to get it in AutoPano (but automatically stitched).

With your rotation (near the equator) rectangles stay rectangles (easier to stitch), but you are changing the panorama! It's totally different now. Too dirty trick — to change a problem that you can't solve to another one, that you can :)

So, basically, yes, I was trying a "limit case" — when pieces come close to the pole.

(*) some computer problem, so I lost yesterday the trial download of AutoPano, so I'm showing in Hugin.

no avatar
Nanard
Member
 
Posts: 150
Likes: 0 post
Liked in: 3 posts
Joined: Sun Jul 29, 2007 11:00 pm
Location: France
Info

Re: Synthetic test, with features. How to make it work?

by Nanard » Wed Aug 01, 2018 9:54 am

What you see at the 'south pole' is absolutely normal and is the result of the pano projection.
You get the same distortion at the left and right sides of a horizontal pano, depending on the angle and projection mode.

"With your rotation (near the equator) rectangles stay rectangles (easier to stitch), but you are changing the panorama! It's totally different now. Too dirty trick — to change a problem that you can't solve to another one, that you can"

I disagree totally. Horizontal panos get distorted as well, depending on the angle. I am not changing the pano, after stiching I rotate it again and get a vertical pano ! A trick is not dirty if it helps solving a problem !


Some basic concepts about panoramic projections (from Autopano documentation) :
Planar or rectilinear projection
Pros:
Good choice if the angle is low, recommended for architectural shots because it is the only mode that does not curve lines that are deemed to be straight lines.
Cons:
Can be used only if the view angle is less than 180 degrees along the diagonal of the image, practice shows that beyond 90 degrees the stretching of the image is already visible along the edges and in the corners.
Beyond 120 degrees, the result is no longer acceptable, because the stretching of the image causes an unpleasant loss of sharpness.

And see also about panoramic projections, you see the distortions :
https://www.cambridgeincolour.com/tutor ... ctions.htm

no avatar
pen_or_rhino
New member
 
Topic author
Posts: 9
Likes: 0 post
Liked in: 0 post
Joined: Mon Jul 30, 2018 12:14 pm
Info

Re: Synthetic test, with features. How to make it work?

by pen_or_rhino » Wed Aug 01, 2018 10:46 am

Ok, looks like we argue about what the problem is :)

If we want to stitch the above 7x1 pano — you right. The trick is not dirty. My wrong. It is totally correct and valid.

If we want to understand capabilities and shortcomings of an algorithm — it is dirty! :)

no avatar
pen_or_rhino
New member
 
Topic author
Posts: 9
Likes: 0 post
Liked in: 0 post
Joined: Mon Jul 30, 2018 12:14 pm
Info

Re: Synthetic test, with features. How to make it work?

by pen_or_rhino » Mon Aug 13, 2018 3:45 pm

Hello again! Actually, I realized I have here a further question.

I tried to load 2 first pieces, it auto-detected, then I added 3d, and manually moved it below (where it should be):
pano3-b.jpg

Notice that it's larger than needed! This is the problem.

Here's what happens, as I see it, very simple. The piece was "smaller" size at non-zero pitch when I "shoot" it. When loading, it was pasted as-is to equator, without proper pre-shrinking = without taking non-zero pitch into account. Then, during the manual move, it gets enlarged to the (very) wrong size.

Am I right in my understanding? Can the original shooting pitch be specified? Where?

Does Autopano optimize with respect to this "shooting pitch"? (may be silently, without showing it to the user?)

Thank you.

no avatar
Nanard
Member
 
Posts: 150
Likes: 0 post
Liked in: 3 posts
Joined: Sun Jul 29, 2007 11:00 pm
Location: France
Info

Re: Synthetic test, with features. How to make it work?

by Nanard » Mon Aug 13, 2018 5:18 pm

You have cut several images out of a single image.
Questions :
1) What is the (vertical) angle of view of the original single image ?
see : https://en.wikipedia.org/wiki/Angle_of_view
2) Same question for the pieces ?
as calculated by AutoPano

Is it right or not ?

Bernard

no avatar
pen_or_rhino
New member
 
Topic author
Posts: 9
Likes: 0 post
Liked in: 0 post
Joined: Mon Jul 30, 2018 12:14 pm
Info

Re: Synthetic test, with features. How to make it work?

by pen_or_rhino » Tue Aug 14, 2018 8:35 pm

1) There's no answer: it's the size you want. For separate pieces you specify the FOV in the settings (double click in the FOV field in the "Edit pano" window; Autopano then auto-calculates focal length).

2) Not sure where to find it out. In "Edit pano" window it currently writes FOV=35.81 (on the screenshot above you can see 35.92, close). But it's horizontal I believe. In the main window (where you have the "Edit pano" button) it writes 52.64 x 53.60 — for the whole 3-piece pano. Which I find believable for the box, surrounding the pano (some areas inside the box will be un-covered by pieces).

I am fine with the Autopano's guess of FOV.

I don't see why you ask me these questions. If they are a hint for me to the solution, I don't get it (and would ask for a more direct answer, not a hint, please).

no avatar
pen_or_rhino
New member
 
Topic author
Posts: 9
Likes: 0 post
Liked in: 0 post
Joined: Mon Jul 30, 2018 12:14 pm
Info

Re: Synthetic test, with features. How to make it work?

by pen_or_rhino » Mon Aug 20, 2018 3:45 pm

So, could the author, or someone experienced/qualified enough comment my concern please?

Once again, briefly: I believe that omission of setting for the original shooting pitch technically is a mistake, because it leads to wrong size and shape of the piece.

My apologies for bumping the thread. Thanks.

UPDATE:
I implicitly meant equirectangular projection (when pieces are in rectilinear, no problems). There's a simple workaround then: put your piece in large enough rectangle so that when pasted at equator, the piece will be pitched where needed. So no problem in this case either. So no need for author's attention. Everything (looks like) makes sense to me now. Thanks.


Who is online

Users browsing this forum: No registered users and 8 guests