Hugin Parameters for Control Point Detectors dialog
This page gives a description of all options. Control point detector parameters site gives an overview about the settings for the different control point detectors and for different use cases.
You should give your parameter-set a meaningful name, because this name is shown in the GUI on the Photos tab.
With this setting all images are transferred at once to the control point detector. This was the default setting of Hugin versions before 2010.2.
This is the same procedure as All images at once with some modifications for using autopano by Alexandre Jenny (only Windows and Linux, there is no MacOS version).
This setting is intended for panoramas with stacks. In the first step it selects the image with the median exposure of every stack. Then it tries to match these median exposed images. After this step all image stacks should be connected.
Then it tries to connect the images of every stack by using the program given under Detector for stacks.
When your panorama does not contain stacks, it's does the same as All images at once.
As the name implies this setting is for multi-row panoramas. But it works also for single row panoramas. It uses a heuristic approach to speed up and optimize the detector step.
The following steps are executed:
1.) Try to connect all consecutive images. The result will be a project with one or more groups of connected images. With luck each of these groups will correspond with a row of the panorama.
2.) Then take the first and last images from each group and match them all together in one go. With luck there will be a single group of connected images in the project.
If you selected Try to connect all overlapping images on the Advanced tab, it tries to connect all overlapping images. This happens only if the steps above were successful. Only then the following steps are executed:
3.) Optimize panorama (it optimizes only yaw and pitch) to get a rough position of all images.
4.) Now try to link all overlapping images which are not yet connected.
This setting combines Panorama with stacks and Multi-row panorama. It works like Panorama with stacks, but for connecting the median exposed images it uses the algorithm described under Multi-row panorama.
This setting works only if the rough positions of the images are defined in the project. It tries to link all overlapping images.
If the advanced option Work only on image pairs without control points is selected (default), it skips all image pairs which are already connected by Control points. Otherwise it creates also control points for already connected images.
On the notebook you can give the paths and parameters for the individual programs.
Here you can give the program name (inclusive path) and the parameters for the control point detector. You can use a one-step setup. In this you invoke the control point detector with a list of image files and it returns a project file with the found control points. If you use cpfind with the --cache parameter you can specify a cleanup argument to delete the generated keypoints file at the end of the detection run.
A further speed up of the detection phase (especially for the multi-row panorama settings) can be achieved by using the two-step detectors approach. Currently only autopano-sift-C and cpfind support this approach. In this case you have two separate programs: a feature descriptor and a feature matcher. The feature descriptor extracts the features from the images and stores a keyfile in the temp directory. Then the feature matcher reads these keyfiles and tries to match overlapping images. At the end of the detection phase the temporary keyfiles are deleted. You have to give the program name and arguments for both programs. A big speed up can be achieved for the multi-row and prealigned settings.
Here you can define the control points detector which works on stacks (in setting Panorama with stacks and Multi-row panorama with stacks only). You can use e. g. Align image stack for aligning the stack. If you used a tripod, you can leave this field empty. Then the detector does not run on the stacks. In this case you can use linking of the image positions on the Hugin Images tab[*].
Typical tweaks might be to:
- Set --noransac for autopano-sift when using non-rectilinear Projection input images.
- Set --size for autopano-sift which by default downsizes images to 700 pixels before matching.
- Some operating systems have problems supplying more than 120 or so photos at one time to these external tools, with autopano-sift-C 2.5.1 or later you can use these alternative arguments to specify input via a project file: --maxmatches %p %o %s
- MacOS only: If you are using control point detector which are contained in the bundle, give only the program name (without path). Hugin tries to find the program then inside the bundle.