Nick Fan wrote:
Any one has a comparison of performance between the Merlin and other more precise motorized head ? Is the stitching speed affected in regard to "creation of templates for automated detention and stitching"?
There is 3 parts in this question :
1. Comparison of panorama heads ( more or less precise )
2. How a motorized panorama heads is used in autopano
3. Templating ( which is a bit separate because templating works also without a motorized panorama head )1. Panorama head comparison :
There are several kind of course, more or less accurate, that can handle more weight ( means longer focal ).
For autopano, we can separate them in two categories :
Type 1 - Panorama heads that creates a log file with shooting position ( merlin, clauss, pixorb )
Type 2 - Panorama heads that don't create any log file ( gigapan )
( I don't know how to classify other panorama head )
Accuracy can change between panorama head, but it's not important in fact.2. How a motorized panorama heads is used in autopano
Let's first talk about the first type of panorama head with a log file.
We parse the log file and extract positions. By using image focal and location, we can calculate overlapping zone and thus we can create a list of pictures pair that should be linked. We do detection on them. Some will have control point, some won't ( becaus full blue sky ).
The optimizer did it's job by combining both information : CPs and approximate location given by the head. So even unlinked pictures will have directly a quite good location.
In this phase, I've noticed a bit shift with some panorama head which gives for example a value and the calculation gives another value. I do'nt ge the time yet to give a general trend.
The second kind of panorama without log file is handled this way.
We only know how the shoot has been done, but we don't have any location. With the structure in the shooting, we can have a potential list of pair of image linked too. That's easy. We do detection on them. The optimizer do a standard optimization without coping with unlinked picture. The remaining unlinked picture are then patched with the knowledge of the structure in the shooting stage ( we don't have the location, we have relative information : this image is on the top of this one which is linked : for example ). It's harder to patch but it works many time.3. Templating
For the moment, templating in autopano works this way :
You have a panorama that you find a great candidate as a standard kind of shooting you are used to do. Right click on save and you'll get a "save as template". This button saves the .pano in a special folder in the autopano installation folder called template. The name of the .pano is the name of the template.
Let say you have a typical 6+1 fisheye job ( 6 in a row at -15° and the nadir, always in this order ).
In a group, when you already have this setup, you just say in the group setting, use the same template. What does it do :
It will use the links in the template to prepare a list of possible candidate in the new panorama. In the 6+1 fisheye case, it will prepare ( 1-2, 2-3, 3-4, 4-5, 5-6, 6-1, 1-7, 2-7, 3-7, 4-7, 5-7, 6-7). With the list, the detection starts and try to find some relation. So with the template you will not be able to find a link between 2-4 for example because it doesn't exist in the template. But even if a link between 1-2 was found in the template doesn't mean it will found one in the current group. We don't transfer CP yet from templates, we prefer to give a higher priority to image analysis.
That's the current implementation. In the future, this solution will give us many more possibilities because we rely on the full .pano as a template :
- we could transfer CPs,
- we could transfer distortion,
- we could transfer color correction.
That's planned too, but after the final 2.0 release. We prefer to concentrate on the creation of a strong background that has many potential evolution than doing directly everything on a weak base.