Fig. 2.
The patch-swap operation. (a) Find the 1×1 boundary patch (most similar) by minimizing d(pi,qj). (b) Replace pi with qj directly as in [25]. (c) Refine p1 with its neighboring patches and the most similar patch with an soft attention scheme. (d) Refine p2 based on the previously refined p1. We note that this patch-swap operation is carried out in the FCN-based CTE’s feature space.