Page 1 of 1

celling problems

PostPosted: Mon Jun 13, 2011 10:05 am
by tbabik
Hello to all users of kolor autopano products.

I've got problem with stiching celling.
The celling got complicated structure and i've try plenty of different control points edit without perfect result.
Mayby you've got some ideas how to deal with that (please help)
Of course the last level of that job is post processing in PS, but it's killer work.
Below are some screen shots to show you what the problem is.
Looking forward for some help :)

Cheers
Tomek

(hardware used to do this pano was: canon 7d + samyang 8mm + manfrotto 303sph, 6 pic taken + zenith and nadir)

PostPosted: Mon Jun 13, 2011 4:25 pm
by klausesser
I´d spontaneously say: the alignment of camera/lens and NPP isn´t absolutely correct. I realized that even 2 mm incorrectness produces errors.
You might try PTGui - it handles such strong geometrical patterns better sometimes. I don´t think the Samyang is a good lens for such structures.

best, Klaus

PostPosted: Mon Jun 13, 2011 6:14 pm
by DrSlony
Try Ransac model: homography.

Also try these options in "Render":
Multiband -4
Diamond weighting OFF
Remove ghosts ON

PostPosted: Mon Jun 13, 2011 7:58 pm
by tbabik
//klaus
thank you for answer, i can't do those pic one more time so i need software solution of this problem.
What do you think about different configuration that i can use in future: canon 5dm2 + 14mm L + manfrotto 303 sph, maybe it will be better for that kind of work,

//DrSlony
thank you for answer, i'll try those settings tommorow and let you know

PostPosted: Mon Jun 13, 2011 10:20 pm
by gkaefer
perhaps setting manually CPs on the overlapping images at these problem zones may help...
Georg

PostPosted: Tue Jun 14, 2011 8:09 am
by tbabik
i've try that solution, without good results, thanks

PostPosted: Tue Jun 14, 2011 8:18 am
by tbabik
//DrSlony
i've try your settings, unfortunately the same problem

PostPosted: Tue Jun 14, 2011 10:25 am
by mediavets
tbabik wrote://DrSlony
i've try your settings, unfortunately the same problem

Did you say it was not possible to reshoot?

If you cannot reshoot - is the ceiling really worth viewing anyway? It's not the Sistine Chapel is it? Perhaps omit the zenith image and just limit the FOV in the panorama viewer?

PostPosted: Tue Jun 14, 2011 1:19 pm
by tbabik
Yes it's impossible to reshoot.
The celling is very important for the designer who made this interior , so i can't limit the FOV.
Thanks anyway.

Tomek

PostPosted: Tue Jun 14, 2011 2:08 pm
by mediavets
tbabik wrote:Yes it's impossible to reshoot.
The celling is very important for the designer who made this interior , so i can't limit the FOV.
Thanks anyway.

Tomek

If you have Autopano Giga then try shooting a different scene with of lot of good features for control point detection using the same shooting pattern, when you have achieved a really good stitch then save as a template and then use that template with your other troublesome pano.

It may improve the alignment.

PostPosted: Tue Jun 14, 2011 2:09 pm
by mediavets
tbabik wrote:What do you think about different configuration that i can use in future: canon 5dm2 + 14mm L + manfrotto 303 sph, maybe it will be better for that kind of work

The Canon 5D2 with the Canon 15mm fisheye is a good combination for shooting spherical panos.

When shooting interiors like this with the canon 7D and the Samyang 8mm fisheye try shooting two rows of 6-around and omit the zenith shot.

I find this pattern works very well for interiors using a Ninoknn DX body and the Nikkor 10.5mm fisheye with a Nodal Ninja 5L head, it seems to overcome these alignment issues between a single zenith and a main row of 6-around around.

PostPosted: Wed Jun 15, 2011 9:54 am
by GURL
klausesser wrote:I´d spontaneously say: the alignment of camera/lens and NPP isn´t absolutely correct. I realized that even 2 mm incorrectness produces errors.

This is not well known but the proof is easy: if the NPP moves by 2mm between two adjacent shoots and the most distant parts of the subject (where 2 mm correspond to a pixel fraction only) are perfectly stitched then the error size will be of 2 mm on nearby subject parts.

My guessi is that the holes in the ceiling are much larger than 2 mm and that the stitch could be improved by deleting useless links and replacing wrong CPs (provided NPP adjustment is in the usual error range.)

BTW, I would try the Multiple viewpoints optimization setting first...

PostPosted: Wed Jun 15, 2011 4:14 pm
by klausesser
GURL wrote:My guessi is that the holes in the ceiling are much larger than 2 mm and that the stitch could be improved by deleting useless links and replacing wrong CPs (provided NPP adjustment is in the usual error range.)

Hi Georges!

Runing exessive tests with Josef´s new Head i realized that 2mm does make a difference with nearby objects - whatever they are: a table´s rim, the hose of a vacuum cleaner, a nearby standing tripod and so on.
In addition there´s the "traditional" weakness ;) of APG dealing with strong geometrical structures - i encountered that problem often and usd PTGui instead in such cases . . which effortlessly stitched the very same image-set perfectly (but has other shortcomings compared to APG).

"The books" tell a NPP to camerascrew-distance of 85mm for the 5D2 and 15mm fisheye. Starting from that value i ended up at 80mm for a perfect stitch. 2 mm of missing the lateral setting caused minor, but clearly visible errors with the 15mm inside a room.

At the moment i´m critically testing the 35mm - didn´t get a perfect stitch yet.

You see: heavy editing should work nevertheless. But that´s not what you really want when producing panos commercially - it takes too much time adding to the anyway sophisticated processing we do as example.

So my advice: the effort you invest in a perfect alignment definitely pays off in the end! :cool:

best, Klaus

PostPosted: Mon Jun 20, 2011 10:25 am
by tbabik
Thank you all for your advices.

cheers
Tomek