I am wondering about SIFT/SURF and Bundle adjustment used by Photoscan.
As far as I can tell these 2 steps are both on the CPU and both multi-threaded.
Are you guys looking into having an option to move theses things onto the GPU to make them many times faster?
SIFTGPU
http://www.cs.unc.edu/~ccwu/siftgpu/MultiCoreBundleAdjustment
http://grail.cs.washington.edu/projects/mcba/I am curious as for example SIFT seems to pretty slow in Photoscan, but I guess that mainly because it is extracting a lot of points using High as value.