Difference between revisions of "Cpclean"
(Added long command line switch available with Hugin 2016.0) |
|||
| Line 4: | Line 4: | ||
===Step 1=== | ===Step 1=== | ||
| − | optimises each pair | + | optimises all image pairs, calculates for each pair mean and standard deviation sigma and then deletes all control points with an error bigger then mean+n*sigma. |
===Step 2=== | ===Step 2=== | ||
| − | optimises the whole project together using more conservative assumptions and deletes another round of points. If there is more than one image chain present this step is not done. | + | optimises the whole project together using more conservative assumptions, calculates then mean and standard deviation sigma and then deletes another round of control points with error bigger then mean+n*sigma. If there is more than one image chain present this step is not done. |
===result=== | ===result=== | ||
| Line 20: | Line 20: | ||
===Options=== | ===Options=== | ||
| − | Vary the threshold with the <tt>--max-distance</tt> parameter (defaults to 2, i.e. points with a distance error greater than | + | Vary the threshold with the <tt>--max-distance</tt> parameter (defaults to 2, i.e. points with a distance error greater than mean + 2 * sigma are deleted). When decreasing this factor to 1, then cp with error bigger than mean+1*sigma will be removed, also removing more control points: |
cpclean --max-distance=1 --output=output.pto project.pto | cpclean --max-distance=1 --output=output.pto project.pto | ||
Revision as of 16:31, 20 May 2022
cpclean is a command-line tool that implements the 'bad' control points pruning available within the Hugin GUI. It is similar conceptually to apclean, but with some extra functionality. Both tools use a statistical filter to remove points with an error above a statistical threshold.
The idea is to remove all non-credible control points. To do this it uses a two step process to remove bad control points.
Step 1
optimises all image pairs, calculates for each pair mean and standard deviation sigma and then deletes all control points with an error bigger then mean+n*sigma.
Step 2
optimises the whole project together using more conservative assumptions, calculates then mean and standard deviation sigma and then deletes another round of control points with error bigger then mean+n*sigma. If there is more than one image chain present this step is not done.
result
The .pto project file is returned unoptimised, the only change is that there are less control points, this means the project doesn't have to be optimised before you start.
By using this optimisation step cpclean is able to look not just at the current project settings and find 'bad' points, but is able to find points that are 'bad' under a range of circumstances.
usage
A typical usage would be:
cpclean --output=output.pto project.pto
Options
Vary the threshold with the --max-distance parameter (defaults to 2, i.e. points with a distance error greater than mean + 2 * sigma are deleted). When decreasing this factor to 1, then cp with error bigger than mean+1*sigma will be removed, also removing more control points:
cpclean --max-distance=1 --output=output.pto project.pto
Do only pairwise optimisation (skip step 2):
cpclean --pairwise-checking --output=output.pto project.pto
Do only whole panorama optimisation (skip step 1):
cpclean --whole-pano-checking --output=output.pto project.pto
Use current image positions for calculation of distance errors (skip optimisation of whole panorama):
cpclean --whole-pano-checking --dont-optimize --output=output.pto project.pto
By default line control point are ignored by the calculations for step 2. With --check-line-cp you can also include these control points into calculation
cpclean --whole-pano-checking --check-line-cp --output=output.pto project.pto
cpclean is based on ptoclean.