Skip to main content
. 2021 Sep 20;122:108341. doi: 10.1016/j.patcog.2021.108341

Fig. 5.

Fig 5

Segmentation is desired to be rotation equivariant. If the rotation is performed on an input image, the ground truth mask should be rotated in the same manner. After the CNNs, the generated output from the rotated image should be the same as the rotation of the original output, which is generally different in fact.