I'm using ceres to solve a multi-view bundle adjustment problem. Besides the classical reprojection-terms I also added a relative movement term for known camera motion.
Since the movement of the camera is predefined and controlled, I would like to give this constraint a higher weight (ask the optimization to give it a higher priority).
At the moment I use ScaledLoss like this:
ceres::LossFunction *motion_loss = new ceres::CauchyLoss(1.0);
ceres::LossFunction *scaled_motion_loss = new ceres::ScaledLoss(motion_loss, 1000., ceres::TAKE_OWNERSHIP);
[...]
problem.AddResidualBlock(cost_function, scaled_motion_loss, parameter_blocks);
Since I am completely inexperienced in using the ceres-solver, I would like to know if this way is reasonable, or if there is a better way.
