Stitching  

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
Boley
Member
 
Topic author
Posts: 31
Likes: 0 post
Liked in: 0 post
Joined: Tue Oct 15, 2013 9:13 am
Info

Stitching

by Boley » Tue Aug 05, 2014 11:59 pm

I was wondering if someone can explain why I would be getting an error in the following workflow.

I have stitched a 6 shot pano with a full frame 16mm fisheye and the stitching is pretty good at 3.03 RMS. When I select preview and enlarge to 100% in APG I cannot detect any stitching errors. When I render it as a 16 bit PSD file I find when I open it up in Photoshop that there are several stitching errors as you see in the image below.

I have tried different blending options but come up with the same error. Does anyone have a reason why this would be happening.

Stitching.jpg


Regards

Robert

no avatar
Boley
Member
 
Topic author
Posts: 31
Likes: 0 post
Liked in: 0 post
Joined: Tue Oct 15, 2013 9:13 am
Info

Re: Stitching

by Boley » Wed Aug 06, 2014 10:43 pm

OK I think I found what the problem was and I am not sure if it is a bug or not. I will share my findings in case others have had the same frustrating issues.

After my last post I continued trying out different things and thought since the problem did not show up in the preview only in the final render a good place to start was in the render set up. First I tried different interpolator settings but with no difference. I then played around with different cutting and blending options but that did not fix it either.

I then played around with slider under the cutting setting (which I have no idea what it is supposed to do) and slid it all the way to the far right toward the "Long focal" end and bang it fixed the problem after I did a render. Just to make certain this was the fix I reset the slider back to the far left (towards Ghost) and the problem reoccurred after I rendered.

I then proceeded to play with various slider positions and found that it worked toward 3/4 of the way toward the far right (see image below). I am really pleased I have found the problem and I am yet to try it on other images that I have had similar issues with, mind you I have had this problem occur on the rare occasions as I do a lot of panos and always shoot with the same set up.

This issues raises more questions than answers:
1. Why does this error not show up in the preview regardless of where this slider is positioned
2. Does this slider only work when rendering in which case why does it appear in the preview settings
3. Is this a bug

Render setting.jpg


Interested to hear from anyone if they have had any such experience, especially someone from the Kolor team would be great.

Thanks

Robert

User avatar
klausesser
Member
 
Posts: 8836
Likes: 5 posts
Liked in: 64 posts
Joined: Mon May 22, 2006 12:18 am
Location: Duesseldorf, Germany
Info

Re: Stitching

by klausesser » Thu Aug 07, 2014 1:35 am

Boley wrote:OK I think I found what the problem was and I am not sure if it is a bug or not. I will share my findings in case others have had the same frustrating issues.

After my last post I continued trying out different things and thought since the problem did not show up in the preview only in the final render a good place to start was in the render set up. First I tried different interpolator settings but with no difference. I then played around with different cutting and blending options but that did not fix it either.

I then played around with slider under the cutting setting (which I have no idea what it is supposed to do) and slid it all the way to the far right toward the "Long focal" end and bang it fixed the problem after I did a render. Just to make certain this was the fix I reset the slider back to the far left (towards Ghost) and the problem reoccurred after I rendered.

I then proceeded to play with various slider positions and found that it worked toward 3/4 of the way toward the far right (see image below). I am really pleased I have found the problem and I am yet to try it on other images that I have had similar issues with, mind you I have had this problem occur on the rare occasions as I do a lot of panos and always shoot with the same set up.

This issues raises more questions than answers:
1. Why does this error not show up in the preview regardless of where this slider is positioned
2. Does this slider only work when rendering in which case why does it appear in the preview settings
3. Is this a bug

Interested to hear from anyone if they have had any such experience, especially someone from the Kolor team would be great.



Hi Robert!

This underlines my own experiences: in my eyes - of course maybe i´m wrong - it indicates that a RMS of 3,03 simply is NOT good enough when you zoom in.

I realized i need a RMS between 1 and 2, preferably RMS 1 or even less for definitely having not even minor stitch-errors.

To me the function of the optimizer´s settings is absolutely obscure regarding what they do and how they can be used optimally.
I don´t realize any effect by using them.

Klaus

no avatar
Boley
Member
 
Topic author
Posts: 31
Likes: 0 post
Liked in: 0 post
Joined: Tue Oct 15, 2013 9:13 am
Info

Re: Stitching

by Boley » Thu Aug 07, 2014 4:56 am

Hi Klaus, I agree with you 100% in regards to the RMS but my greatest concern is the fact that if there are stitching errors no matter how small they should show up in editing and not be seen after you render the pano otherwise you are just guessing and wasting valuable time, which, for people who do this for a living means money. In APG at 100% zoom I did not see these errors until I rendered the pano and zoomed in at 100% in PS.

Can someone from Kolor provide some light on this, perhaps there is a logical explanation.

Regards

Robert

no avatar
toto-foto
Member
 
Posts: 35
Likes: 0 post
Liked in: 0 post
Joined: Fri Dec 14, 2012 7:48 am
Location: Moscow
Info

Re: Stitching

by toto-foto » Thu Aug 07, 2014 11:30 am

Hi, Robert!
I have the same problem even if RMS is 1.97... And I used to regard it "as is" and always had a lot of retouching.
But this trick - moving the slider closer to "long focal" - solves the problem... It's great!!! Thank you for investigation!
scr.jpg

Best Regards
Anton

User avatar
klausesser
Member
 
Posts: 8836
Likes: 5 posts
Liked in: 64 posts
Joined: Mon May 22, 2006 12:18 am
Location: Duesseldorf, Germany
Info

Re: Stitching

by klausesser » Thu Aug 07, 2014 11:31 am

Boley wrote: . . . otherwise you are just guessing and wasting valuable time, which, for people who do this for a living means money.


Hi Robert!

Right!

Klaus

User avatar
klausesser
Member
 
Posts: 8836
Likes: 5 posts
Liked in: 64 posts
Joined: Mon May 22, 2006 12:18 am
Location: Duesseldorf, Germany
Info

Re: Stitching

by klausesser » Thu Aug 07, 2014 11:34 am

toto-foto wrote:But this trick - moving the slider closer to "long focal" - solves the problem... It's great!!! Thank you for investigation!


Hi Anton!

I´m sure there are some tricks to solve this. But i mean it´s up to Kolor providing a
real good description how to do it.

What i read in what they call "documentation" isn´t of any help at all regarding this issues.

Klaus

User avatar
klausesser
Member
 
Posts: 8836
Likes: 5 posts
Liked in: 64 posts
Joined: Mon May 22, 2006 12:18 am
Location: Duesseldorf, Germany
Info

Re: Stitching

by klausesser » Thu Aug 07, 2014 12:26 pm

Robert, Anton - maybe you can find some vital information here:

http://www.autopano.net/wiki-en/action/ ... _RMS_value

I can´t judge what´s vital or not - i can only guess.

Klaus

no avatar
Boley
Member
 
Topic author
Posts: 31
Likes: 0 post
Liked in: 0 post
Joined: Tue Oct 15, 2013 9:13 am
Info

Re: Stitching

by Boley » Thu Aug 07, 2014 1:29 pm

Hi Anton, I am glad that I am not the only one to have this problem but even happier that it was able to help you solve the problem I understand your frustrations and it's been pretty much like this for the whole time I have been trying to learn this software.

Klaus, I can understand your frustrations and throughout these forums it's echoed by so many other users that the documentation is inadequate for such a great and complex software.

What I find most disappointing is that Kolor have not responded at all, not to this subject or any others I have posted. I think that it is their obligation and in their interest to do so and have satisfied customers. It should be driven by them, it's unlike any other forum that I have been involved in.

I guess all we can do is to stumble along until we get it right.

Robert

User avatar
AlexandreJ
Kolor Team
 
Posts: 5987
Likes: 7 posts
Liked in: 10 posts
Joined: Mon Nov 14, 2005 4:56 pm
Location: Francin, France
Info

Re: Stitching

by AlexandreJ » Thu Aug 07, 2014 4:15 pm

RMS value comprehension is complicated. But there are a few rules that you need to follow:
* Do not rely too much on it as it can gives you a false impression of good. As I have said several timse, it is possible to have a RMS of 0 just by taking 2 images, put randomly a single CP on these 2 images and optimizing.
So RMS is nice as it says globally the trend, but not everything a stitch.
* (A) What is important is does it move or not when inspecting the panorama with the image tool. I'm used to flying over the panorama while mouse wheeling to switch between images for the whole panorama. Doing that, normally, nothing should pop and stay at the same position. If at any point, something pops, it may be a ghost or it may be some lines that are not aligned.
* On that pair of images, I then look at the CP and only at this moment. Globally, there are only 2 types of mistakes there. Mistake 1: repeating patterns, it fooled the software, so just remove the cp and put then on the right patterns.
Mistake 2: the cp are cluttered in one zone and not over the whole overlapping zone. Just correct that by adding cp using the area tool.
* Re-optimize and it should be good.

If it is not good, then, it is needed to have a second analysis:
- Where all shots taken at the same focal length or not ?
- Did I change the focus ?
- Did I use several cameras ?
- Did I hit accidentally the zoom (which could change slightly the focal, but not the EXIF written ... ) ?
- Etc.
All these factor affect the scope part of the optimizer.

Which brings me back to the first question in this thread. Why a difference ?
- The default view is a dirty view with linear blending for making the interface responsive. It can smooth some issues.
- If you switch to realtime preview, then, you use the same engine as the rendering and the issues can appear (if they exist). That's why I use (A) always when inspecting a panorama.
- For us, it is always complicated to get a real answer without the real images. That's why comment on RMS are quite rare as it is hard to give the real reason without them. It could come from a metadata issue, or one of cause quoted before, or even because you use photomatix before.

User avatar
klausesser
Member
 
Posts: 8836
Likes: 5 posts
Liked in: 64 posts
Joined: Mon May 22, 2006 12:18 am
Location: Duesseldorf, Germany
Info

Re: Stitching

by klausesser » Thu Aug 07, 2014 6:34 pm

AlexandreJ wrote:RMS value comprehension is complicated. But there are a few rules that you need to follow:
* Do not rely too much on it as it can gives you a false impression of good. A I said several time, it is possible to have a RMS of 0 just by taking 2 images, put randomly a single CP on these 2 images and optimizing.
So RMS is nice as it says globally the trend, but not everything a stitch.
* (A) What is important is does it move or not when inspecting the panorama with the image tool. I'm used to fly over the panorama while mouse wheeling to switch between images for the whole panorama. Doing that, normally, nothing should pops and stay at the same position. If at anything something pops, it may be a ghost or it may be some lines that are not aligned.
* On that pair of images, I then look at the CP and only at this moment. Globally, there are only 2 types of mistakes there. Mistake 1: repeating patterns, it fooled the software, so just remove the cp and put then on the right patterns.
Mistake 2: the cp are cluttered in one zone and not over the whole overlapping zone. Just correct that by adding cp using the area tool.
* Re-optimize and it should be good.


Alexandre - everything you say is right. BUT: i guess it´s related to "normal" resolutions.
"Flying" over the panorama and watching the movements i did for years - as long as i shot i a range of 112mpx with a 5D2 an 15mm fisheye = 7 shots.
But "flying" over the panorama when you have 45 shots = 1,2Gpx or 214 shots = 6 gigapixels it is impossible to verify very small moves in a range of 2 or 3 pixels.

And that´s the point which i try to underline: A stitch-error which you can see by "flying" over the low-res stitch you realize immediately.
A stitch-error of 2 or 3 pixels in a 700mpx stitch you most unlikely wouldn´t see at all by "flying" over in the editor of rather low resolution.

So when the RMS show 3 - in my experience it indicates a stitch-error you definitely will see in a 700mpx stitch when you zoom in 100% . . . while
you most likely will NOT see it in a 112mpx stitch.

And THAT´S the problem.

Editing the images pair-wise in the CP-editor by manually deleting RMS over 2 and each time fast-optimize in between provides me a RMS of 1 or lower.
And the stitch is perfect - even zoomimg in to 100% i cannot find a single stitch-error.

THAT´S the way i needs to be!
But achieving this value of 1 or less and having a perfect stitch takes very much time by doing it manually on image-pairs.

AlexandreJ wrote:If it is not good, then, it is needed to have a second analysis:
- Where all shots taken at the same focal length or not ?


Yes - of course!

AlexandreJ wrote:- Did I change the focus ?


No - of course not.

AlexandreJ wrote:- Did I use several cameras ?


No - of course not.

AlexandreJ wrote:- Did I hit accidentally the zoom (which could change slightly the focal, but not the EXIF written ... ) ?
- Etc.


No - i always use prime-lenses.

AlexandreJ wrote:All these factor affect the scope part of the optimizer.


Right - but they´re all of no means here.

AlexandreJ wrote:Which brings me back to the first question in this thread. Why a difference ?
- The default view is a dirty view with linear blending for making the interface responsive. It can smooth some issues.
- If you switch to realtime preview, then, you use the same engine as the rendering and the issues can appears (if they exists). That's why I use (A) always when inspecting a panorama.


Which tells us that the realtime-preview´s reslution is too low for judging in cases of 700mpx and higher-res shoots.

AlexandreJ wrote:- For us, it is always complicated to get a real answer without the real images. That's why comment on RMS are quite rare as it is hard to give the real reason without them. It could come from a metadata issue, or one of cause quoted before, or even because you use photomatix before.


I understand that. But what do RMS-values tell us at all as long as we cannot rely on them as a judgable value?

Sorry, Robert and Anton - i didn´t want to occupy this thread . . but i think we share the same issues . . . . which, by the way, also Hans mentioned before
regarding the low resolution of the preview.

Which i tend to agree these days: zooming in to 100% in the preview would make a great difference in judging the stitch!
Zooming in actually does not show subtle stitch-errors @100% in a 700mpx stitch . . .

Klaus

no avatar
Boley
Member
 
Topic author
Posts: 31
Likes: 0 post
Liked in: 0 post
Joined: Tue Oct 15, 2013 9:13 am
Info

Re: Stitching

by Boley » Thu Aug 07, 2014 10:42 pm

Hi Alexandrej, I appreciate your input but it still doesn't answer my original question. Why does the stitching error (shown in my posted image) not show up in the real time preview but it does in the final rendered image. If the preview uses the same engine as the render you would expect both images to be identical. For this problem I don't believe that it is in the way the image was taken.

Sorry, Robert and Anton - i didn´t want to occupy this thread . . but i think we share the same issues . . . . which, by the way, also Hans mentioned before regarding the low resolution of the preview.

Which i tend to agree these days: zooming in to 100% in the preview would make a great difference in judging the stitch!
Zooming in actually does not show subtle stitch-errors @100% in a 700mpx stitch . . .


Is it as Klaus suggests that the quality of the preview is not as high quality as it should be to show stitching errors of 1 or 2 px. That's the only reason that I can see for my above issue.

Thanks

Robert

User avatar
klausesser
Member
 
Posts: 8836
Likes: 5 posts
Liked in: 64 posts
Joined: Mon May 22, 2006 12:18 am
Location: Duesseldorf, Germany
Info

Re: Stitching

by klausesser » Fri Aug 08, 2014 11:06 am

Boley wrote:Is it as Klaus suggests that the quality of the preview is not as high quality as it should be to show stitching errors of 1 or 2 px. That's the only reason that I can see for my above issue.


Some time ago i had an arguing with Hans about that topic. I mentioned that a preview working fine @100% must be slowing down the appremarkably - and that a complete test-rendering at lower resolution might be preferable.
In the case of hires shots i was wrong: a test-rendering @25% (to get it quickly) most unlikely could reveal stitch-errors of 2 or 3 pixels. Only a 100% view shows them.

"The other" stitcher application :cool: shows that it´s "detail viewer" can show my 700mpx stitches @100% in render-quality almost in reatime - and much more important: by setting a loupe i can move around a crop in realtime for checking smallest areas in terms of stitch-errors @100% view.

I mean that we either need a relyable RMS-value strategy or/and a 100% preview window in which we can move to an area of interest for checking for stitch-errors of few pixels.

In my eyes Kolor focuses too much on "automatic" more and more. That´s fine for average resolutions - no doubt - but it
definitely can rise problems when you shoot hires indoors.

Alexandre kindly posted a strategy of dealing with the optimizer in terms of achieving lower RMS globally - withou the need of treating each image-pair seperately.

I will try this strategy over the weekend. If it works fine and reliably . . . . then there´s no need for a preview @100% . . . =D :cool:

Klaus

User avatar
klausesser
Member
 
Posts: 8836
Likes: 5 posts
Liked in: 64 posts
Joined: Mon May 22, 2006 12:18 am
Location: Duesseldorf, Germany
Info

Re: Stitching

by klausesser » Fri Aug 08, 2014 11:17 am

Boley wrote: For this problem I don't believe that it is in the way the image was taken.

Sorry, Robert and Anton - i didn´t want to occupy this thread . . but i think we share the same issues . . . . which, by the way, also Hans mentioned before regarding the low resolution of the preview.


I guess that no head in the world can be precise enough to not need ANY kind of optimizing when it comes to hires shoots.
Even the max. offset of 0,036°/10000steps which the Panoneed provides and by using the xml might result in stitch-errors
when you view the 1gpx rendering - which a 35mm on my D800 provides - @100%.

What i forgot to mention: i didn´t deactivate the optimizer-stage in the PW-import module.

I will run the stitch again without automatic optimizing.

We´ll see.

Klaus

PS. - a rising question in such cases also is: how precisely can you really set the NPP regarding indoor hires shots?
I will also research my setting. As i experienced: one millimeter of mismatch causes very visible errors when you view @100% . In that case the optimizer needs to help. IF it can help . . .


Who is online

Users browsing this forum: No registered users and 4 guests