n and the number of pixels in T is N , the computational complexity of the inverse compositional algorithm is O(nN + n3) per iteration and O(n2N ) for pre-computation (performed only once), which is a substantial saving from the O(n2N + n3)-per-iteration Lucas-Kanade algorithm [3]....
the optimal Lucas-Kanade baseline you want to compared with since we use the same stopping criterion, Gauss-Newton solver within the same framework as our learned model. There is no extra bells and whistles, but it may provide a baseline for you to explore the algorithm in various directions...
Bowden. Mutual Information for Lucas-Kanade Tracking (MILK): An Inverse Compositional Formulation. IEEE Trans. Pattern Anal. Mach. Intell., 30(1):180-185, Jan 2008. 3Dowson, N., Bowden, R.: Mutual Information for Lucas-Kanade Tracking (MILK): An Inverse Compositional Formulation. PAMI 30...