Page 1 of 1

APP3 produces stiching errors whereas APP2.6 doesn´t Win7 64Bit

PostPosted: Mon Nov 26, 2012 2:31 pm
by byk
Hello there,
lately I´ve been using the Beta and RC versions of APP3 intensively. The versions run stably and I like the added projections but they won´t stop producing stiching errors in every panorama I´ve tried so far (all multi row, consisting of 18 to 90 pictures each, shot with a panohead). The errors consist in the pictures being combined a bit askew. This forced me to switch back to APP2.6 which stiches the same panoramas just perfectly. I have 32GB RAM which all versions manage to utilize completely, RMS is alway between 1,3 to 2,3. If further information is needed please don´t hesitate to ask.

Best regards,
Christian

PostPosted: Mon Nov 26, 2012 2:45 pm
by gkaefer
welcome to fourm Christian,

can you post some screenshots of these stitching errors?
or even better a download link to a resized set of images so we can try to find better settings in apg 3.0 rc1...
Liebe Gruesse,
Georg

PostPosted: Mon Nov 26, 2012 9:00 pm
by byk
Thanks for the prompt reply! I will upload the wrongly stiched pictures tomorrow.
Christian

PostPosted: Wed Nov 28, 2012 6:08 pm
by byk
Hello again,
Sorry for the delay, I was too busy yesterday.
Out of 18 TIFFs I had APP3RC and APP2.6 render a panorama with the same settings: Detection set to `high´, Mercator projection, Spline64, everything else left at default setting. APP2.6 gains an RMS of 1.59, APP3RC 1.99. I´ve found seven stiching errors in APP3RC´s pano. Depending on the extent of shifting, the askew combinded pictures produce a doubling or smearing effect. The cutouts are not cropped, the first one is always APP2.6, the second one APP3RC.

Christian

PostPosted: Wed Nov 28, 2012 6:11 pm
by byk
The next ones:

PostPosted: Wed Nov 28, 2012 6:13 pm
by byk
#

PostPosted: Wed Nov 28, 2012 6:14 pm
by byk
#

PostPosted: Wed Nov 28, 2012 6:15 pm
by byk
#

PostPosted: Wed Nov 28, 2012 6:16 pm
by byk
#

PostPosted: Wed Nov 28, 2012 6:18 pm
by byk
#

PostPosted: Thu Nov 29, 2012 9:13 am
by ThomasV
Hi Christian,

Could you please tell us if your shots are bracketed ?
From the screenshots, it looks as if the anti-ghost did not run on your pictures.

Thanks,
Thomas

PostPosted: Thu Nov 29, 2012 11:47 am
by byk
Hi Thomas,
no the shots are not bracketed and as I wrote, I left everything except the mentioned alterations at default setting, which includes anti-ghost.

Christian

PostPosted: Tue Dec 04, 2012 9:16 am
by ThomasV
Hi Christian,

To push the diagnostic further, it will be good if you can upload your original pictures and the .pano file on our anonymous ftp (http://www.kolor.com/forum/t766-ftp-server).

Thanks,
Thomas

PostPosted: Tue Dec 11, 2012 3:14 pm
by byk
Hi,
also with APP3.0.0 I get the same stiching errors and even a new bug: When I choose "File>Select Images" only folders are visible, but no files. I have to use the detour via "Browse Folder" to be able to select my pictures. I´m already uploading the requested files, but be patient - the upload takes about 10h.

Christian

PostPosted: Tue Dec 11, 2012 4:04 pm
by ThomasV
Hi Christain,

Thanks for taking the time to upload. I will give you feedback once analyzed.

Regards,
Thomas

PostPosted: Wed Dec 12, 2012 11:17 am
by ThomasV
Hi Christain,

I gave a look at your panorama this morning, and I noticed 2 things.
First of all, I rendered it with 2.6.4 and 3.0 Autopano versions. There are stitching defaults in the 2 (with the default options). In some pieces of the panorama, the 2.6.4 version is better, and in some other pieces, the 3.0 one is.

But the main issue with this panorama is the number of images. As there are paralax defaults, stitching 6 overlapping pictures is quite challenging, and indeed the whole middle row in your panorama is useless. I tried to remove images 7 to 12, and re-optimize the whole panorama, the rms decrease a little bit further, and I didn't notice defaults in the resulting panorama.

Could you please give it a try to see if we conclude the same ?

Thanks,
Thomas

PostPosted: Wed Dec 12, 2012 2:23 pm
by byk
Hello Thomas,
you´re absolutely right, I can confirm that skipping the middle row really does the trick! I have always assumed, that more overlapping means better results - now I understand, that more than 50% overlapping even harms. My other panos with stiching errors are also affected by to much overlapping. I will try your skipping-remedy on them, too, and inform you about the results. Thanks a lot for this fundamental insight!

Greetings from Munich,
Christian

PostPosted: Wed Dec 12, 2012 3:44 pm
by mediavets
byk wrote:I have always assumed, that more overlapping means better results - now I understand, that more than 50% overlapping even harms. My other panos with stiching errors are also affected by to much overlapping.

25- 30% overlap is about 'right' - more is not better.