Computation:
This measures an algorithm’s time-to-completion in MATLAB, and provides an estimate its computational complexity. Although this metric gives a relative idea for comparing algorithms in terms of their complexity, it has the limitation that it cannot leave-out non-computational overhead with various algorithms. For instance, L1 and L2 norm minimization algorithms call external convex optimization packages and image impainting algorithms use functions that are not built into MATLAB. Obviously, these algorithms run slower than other interpolation schemes that use built in MATLAB functions, which are optimized for running in this environment. Hence, the computation metric does not treat each algorithm fairly. However, it is still a useful quick-and-dirty metric for revealing information on the computational complexity of different algorithms. |
