Page 1 of 1

Detecting settings

PostPosted: Sun Jul 08, 2012 9:05 pm
by leifs
I'm struggling with a pano that is supposed to be a piece of cake. But there must be a checkbox somewhere that has to checked/unchecked to get it right. So I ask for suggestions.

case:
gigapano, shot with Oly E-M5 with Canon 400mm using Seitz VRdrive 2
1116 images shot in ORF (raw) transformed to tiff using Lightroom 4.1
APG 2.63 and 3.0alpha gives the same result

the problem:
in adjacent images where there are lots of CP's result in none CP's detected at all
in the example below image 686 and 687 should have a lot of CP's but have none !

anybody ?

leifs

PostPosted: Sun Jul 08, 2012 11:35 pm
by gkaefer
using the v3alpha1 I could misconfigure autopano in any combination. I did not reach to get no single or any less CPs. (in any combination the CPs got created linke in screenshot...)
So I think its not the image. I didnt care EXIF data etc. I created the two "source" images by copy&paste from your screenshot. even so no single problem.

What I think its not normal that this row of images does not have any CPs to images one row bleow (so many images should have enough similar points to detect CPs, the water area its clear that no CPs are detected) . The distance from the rows seems to be equal, so in higher rows CPs are there, but not in the lower...
by testing settings I saw that using the lens disortion like you did brings maybe a corrected image but look to the form of the resulting image (second screenshot the detection result on bottom)... now imagine the same with the row below... I think the overlap of the images are now to small, no CPs can be detected. This does not explain in detail why the left and right image of the same row cant find CPs, but maybe autopano does here fail...

I would use the default gigapixel preset and I would let the lens disortion disabled. autopano takes images and squeezes, stretches and compresses the images to create a panorama so I would really not care about lens disortion on your kind of panorama.

Georg

PostPosted: Sun Jul 08, 2012 11:49 pm
by klausesser
leifs wrote:anybody ?

leifs

Hi leifs!

Sorry for asking - it didn´t come clear to me: did you use your head´s xml-files?

Guess you should hard-link all or some of the red dots to neighboured images and run local opt. each time.

best, Klaus

PostPosted: Mon Jul 09, 2012 10:35 am
by leifs
klausesser wrote:Sorry for asking - it didn´t come clear to me: did you use your head´s xml-files?
best, Klaus

Yes, I did.
That's why the images on the water is nicely put in place :-)
Rendering this as is creates a mess on land, since there are small offsets. Therefore the dots has to be connected, and I would like APG to do it.

leifs

PostPosted: Mon Jul 09, 2012 12:55 pm
by klausesser
leifs wrote:
klausesser wrote:Sorry for asking - it didn´t come clear to me: did you use your head´s xml-files?
best, Klaus

Yes, I did.
That's why the images on the water is nicely put in place :-)
Rendering this as is creates a mess on land, since there are small offsets. Therefore the dots has to be connected, and I would like APG to do it.

leifs

I see! Did you read the little tutorial Alexandre wrote for one of my cases some time ago regarding importing and optimizing:

http://www.autopano.net/wiki-en/action/view/Troubleshooting_Papywizard_Cases

best, Klaus

PostPosted: Mon Jul 09, 2012 2:20 pm
by leifs
I have reread it, but can't find the case to be similar.

Now I have connected some images by hand, and it works fine. Lots of CP's.
I have also tried left-click and run "local geometry analyse" and work fine too.

But why do I have to do it by hand, image for image ? Thats not Auto-pano ! Why doesn't APG find these CP's during the detection, when it easely find them when don by hand ? Each "local geometry analyse" takes me 13 secs and there are a lot of connections to be made. Frustrating.
There must be a better way than this image-by-image-by-hand.

btw: now I'm using APG 2.63
below: the full pano detected, the CP's found by hand, the connections done by hand is blue

leifs

PostPosted: Mon Jul 09, 2012 4:09 pm
by klausesser
leifs wrote:I have reread it, but can't find the case to be similar.

Now I have connected some images by hand, and it works fine. Lots of CP's.
I have also tried left-click and run "local geometry analyse" and work fine too.

But why do I have to do it by hand, image for image ? Thats not Auto-pano ! Why doesn't APG find these CP's during the detection, when it easely find them when don by hand ? Each "local geometry analyse" takes me 13 secs and there are a lot of connections to be made. Frustrating.
There must be a better way than this image-by-image-by-hand.

btw: now I'm using APG 2.63
below: the full pano detected, the CP's found by hand, the connections done by hand is blue

leifs

You don´t need to hard-link all - i realized that linking some images manually where a gap is most likely "triggers" enough of the others.

best, Klaus

PostPosted: Mon Jul 09, 2012 4:33 pm
by hankkarl
If you uncheck "use grid position" and re-optimize (start from a pano you showed above, not from new) what happens to the unlinked pictures?

I suspect that this has to do with CPs getting put on the water instead of on land, and APG then cleaning them out because the grid is prefered.

Georg showed two renders of the images, one was rectangular, the other had severe pincusion. I would expect a rectangular render from two images takein with a 400mm lens. The pincusion was probably caused by bad CPs, and IMO waves give lots of bad CPs.

PostPosted: Mon Jul 09, 2012 4:52 pm
by leifs
The images were shot using the Canon 100-400mm f/4.5-5.6 L @400mm on a crop 2 camera, E-M5, so I would not expect much distorsions.

leifs

PostPosted: Mon Jul 09, 2012 6:54 pm
by leifs
hankkarl wrote:If you uncheck "use grid position" and re-optimize (start from a pano you showed above, not from new) what happens to the unlinked pictures?

this is what happened :-|

leifs

PostPosted: Tue Jul 10, 2012 9:40 am
by AlexandreJ
I don't know the reason, but it seems that calculations don't allow these links to be processed.

One idea : reduce the focal length to 600mm instead to 800mm.

Note : papywizard mode in 3.0 alpha 1 is a bit broken, just use 2.6.3.

PostPosted: Tue Jul 10, 2012 1:49 pm
by leifs
Round #5:
detection quality "high" gives me som more connection, but it takes hours
I connect the important images, mostly by "local geometry analyses"
it look allright, I only need to delete some CP's and optimize.

optimize mess up the hole thing !
dead end. hours wasted. back to square one :-(

leifs

below:
first image: after the optimize
second image: before the optimize

PostPosted: Thu Jul 12, 2012 7:11 pm
by leifs
AlexandreJ wrote:One idea : reduce the focal length to 600mm instead to 800mm.

round #6
600mm. dead end

leifs

PostPosted: Thu Jul 12, 2012 7:21 pm
by leifs
round #7
800mm, no lens correction, standard detection
this is the best APG can do Auto (I'm a bit dissappointed)
I connected the images with buildings, roads etc by hand, one by one.
100 x leftclick, local geometric analyses, next

APG happily find lots of connections and CP's this way. No problem at all.
Why didn't it find them in the first place, Auto ????????

It took me 36 minutes to shoot this 8.3 gigapixel pano and in the end I'm satisfied with the result.
But I'm not happy with all the hours it took to detect it.
With the prime lens, the precision of the VRdrive 2 and the XML it should have been a piece of cake, auto.

here it is
http://www.rundskuer.no/panotour/ulsteinvikframosvarden/virtualtour.html

leifs

below:
a piece of of the pano as detected. lots of missing connections