Table 2.

Single-threaded runtimes on a 2.2-GHz Intel Xeon Silver 4114 CPU, averaged over $n=10$ runs.

Method	Affine (seconds)	Deformable (seconds)
ANTs	777.8 $\pm$ 36.0	17189.5 $\pm$ 52.7
NiftyReg	293.7 $\pm$ 0.5	7021.0 $\pm$ 21.3
Deeds	142.8 $\pm$ 0.3	383.1 $\pm$ 0.6
Robust	1598.9 $\pm$ 0.8	–
FSL	151.7 $\pm$ 0.4	8141.5 $\pm$ 195.7
C2FViTa	43.7 $\pm$0.3	–
KeyMorph	36.2 $\pm$ 2.6	–
VTNb	–	63.5 $\pm$ 0.3
SynthMorph	72.4 $\pm$ 0.8	887.4 $\pm$ 2.5

Errors indicate standard deviations. On an NVIDIA V100 GPU, all affine and deformable DL runtimes (bottom) are ~1 minute, including setup.

^aTimed on the GPU as the device is hard-coded.

^bImplementation performs joint registration only.