Another feature coming soon to Hugin is a new class of intelligent control point improvement. Celeste prunes bad Control Points (CPs) on moving clouds. At about 80% accuracy it does a good enough job to reduce the effect that the bad CPs have on traditional correlation statistics.
It is the result of Tim Nugent’s three months Google Summer of Code project, and I can’t say that I had a lot of work mentoring him. Maybe because of this it does not yet have a CMake build system like all other parts of Hugin, and I only managed to build it in Linux. It wasn’t even that slow on my new low-power Atom box, but who cares about speed on a secondary box sitting next to my workstation and normally used as a test box and music player?
Since I keep my stitching projects on a network drive, the process was quite easy:
- set up the project on my workstation and save it
- access it with Celeste from the Atom box
- issue the single command that runs Celeste
$ celeste -i jpeg.pto
- open the new project file with PTBatcher (see tomorrow’s sneak preview) and stitch it.
The two screenshots above are of a representative image-pair in the project. Out of a total of 20 CP, Celeste has left ten of them. Three happens to be bad and should have been removed. No false positives and a great, though not representative (too small sample) statistics.
Coming to Hugin after the 0.7.0 release.