When providing XML for images, whilst it clearly isn't accurate enough to directly stitch from, and control points need to be generated, it would maybe make sense to use the XML to prevent control points being generated between images that logically could not share any common areas based on the XML.
The XML, along with knowledge of the FoV of the lens could be used to prevent the software from even attempting to create links between images that quite clearly could not be linked.
In a recent pano of a subject that has repeated patterns over a wide area, a lot of links were created that logically would be impossible based on this simple knowledge.
I agree very much with Gerald. And I would even go further: the accuracy of the motors is in the order of 0.1 deg after APG has placed the images according to the XML it should not try to find CP's/optimize moving the images more than +/- x deg and some room for roll. x deg must be decided after some testing, maybe x=1 ? if it can't find CP's: leave the image at the XML position.
Your are both so true. Some notes : - Gddxb : no, the xml is not enough to be able to determine if there is a link or not. We also need the true 35mm focal length. Imagine 2 images separated by 10°. If they are 24mm, they cover each other. If they are 300mm, they don't. So most of the time, when you have such issue where you see link that shouldn't exist, it comes because exif were wrong. - Leifs : yes, that's my idea of the optimization stage too, but I weren't yet able to make it work. So the optimization is still working in another way. This approach has a drawback, it is awfully slow to converge. It took take severals minutes ( 20, 30 ) until it finds the real solution. Compared to the current solution quite always below 1 minute.
So most of the time, when you have such issue where you see link that shouldn't exist, it comes because exif were wrong.
What about forgetting EXIFs? For example: i don´t have EXIFs of my lenses because i use elder Nikon and Zeiss lenses - which are of very high optical quality - on the 5D2. So only camera/sensor-EXIFs exist. Nevertheless it´s sometimes useless to work with the import-module - optimization makes a mess out of an already good looking pattern of images. So sometimes there isn´t any advantage using the xml import at all! Sometimes it work fine with the same configuration.
Seems the importer resp. APG sometimes is unable to definitely decide what to do.
It´s frustrating to use xml import - and having to spend hours and hours afterwards with searching glitches when doing a highres pano! A RMS of around 1,5 or 2 means to have a lot of glitches in a gigapixel. A RMS of 0,8 also means the need of controlling the whole image @100% view to find errors.
So finding and correcting errors in Photoshop can mean to take more time than stitching/rendering the whole panorama.
That´s not really what i expect from working with xml-import . . .
The use of the xml-import and also the use of optimization is a somewhat clandestined operation i mean - you never really know about what the hell you´re doing . . .
Last edited by klausesser (2012-01-09 15:21:18)
If you want something you´ve never had, then you´ve got to do something you´ve never done.