I often spend a lot of time correlating the images to the gpx track, by adjusting the time offset. I go to turns visible in GPX and adjust the offset until I see the turns reflected in the pictures.
Sometimes I found then that my edits deviate for one or two seconds and the images are placed somewhat off.
However, thinking about it, what I am doing there is manual SfM!
So what I am wishing for: Just let me upload the pictures and the gpx separately. Then let the SfM work out individual connected sequences in my long sequence and then use those to exactly align GPX track and pictures.