This work generated significant discussion but in the end we decided to accept it, modulo some revision on your part. As discussed in the rebuttal phase, be sure to do the following: ** Change the algorithm for subspace selection and replace the existing results for the current one. Comment on convergence guarantees (if any) and runtime of the algorithm. ** Include the color transfer application experiments from the rebuttal. As pointed out in the reviews, the main weakness of this work is that it does not include applications of the MK/MI **maps** computed using the algorithm (as opposed to applications of the objective value). If there are other simple applications you can include in the final revision, this would strengthen the work considerably. And of course in your revision be sure to edit responsively to other comments/typos identified in the reviews.