Skip to main content
. 2021 Jul 26;37(8):085006. doi: 10.1088/1361-6420/ac104f

Figure 6.

Figure 6.

The reconstruction quality, as measured on a validation set, of learned proximal gradient methods trained on the CT reconstruction problem with varying orders of the group H. Note that when H is chosen to represent on-grid rotations (i.e. m = 2 or m = 4), the performance is significantly better than for any of the other choices of H.