question about shift 2 in upsampleAlignments #5
Description
dear author，when talking about ‘upsampleAlignments’ function, there is a code like ti1 = np.repeat(np.clip(2 + np.arange(h) // repeatFactor + neighborsMask[:, 0, 0, 0], 0, paddedPreviousAlignments.shape[0] - 1).reshape(h, 1), w, axis=1).reshape(h * w), im just wondering why 2 was used here, what about shift 1? it will be very glad to receiving ur answer!