Is fuse appropriate for doing visual odometry inside the framework? i.e. having fuse optimize the matched keypoints directly? Or is it expected that you will be doing the matching externally and just sending the transformation + covariance to fuse in order to fuse it?
If so, are there examples of such use cases?