Currently loads kitti dataset after processing by VO code (stereo_factors.txt and camera_poses.txt)
Creates graph after N landmarks have been seen, runs it through SmartProjectionFactor
Verification of output is still needed; LM seems to optimize but poses don't seem to change.