Currently our implementation is not very efficient. We allocate an array of ones to create homogenous points before the matrix multiply with the transformation matrix, which is unnecessary. The new implementation simply does points @ R.T + t which is all that is needed.
This is also how other libraries like GTSAM implement this (see here).
Currently our implementation is not very efficient. We allocate an array of ones to create homogenous points before the matrix multiply with the transformation matrix, which is unnecessary. The new implementation simply does
points @ R.T + t
which is all that is needed.This is also how other libraries like GTSAM implement this (see here).