. Author manuscript; available in PMC: 2023 Nov 1.

Published in final edited form as: Med Image Anal. 2022 Sep 14;82:102615. doi: 10.1016/j.media.2022.102615

Table 7:

Average training and inference time for methods used in this work. Note that SyN, NiftyReg, and deedsBCV were applied using CPUs, while LDDMM and the learning-based methods were implemented on GPU. Inference time was averaged based on 40 repeated runs.

Model	Training (min/epoch)	Inference (sec/image)
`SyN`	-	192.140
`NiftyReg`	-	30.723
`LDDMM`	-	66.829
`deedsBCV`	-	31.857
`VoxelMorph`-1	8.75	0.380
`VoxelMorph`-2	9.40	0.430
`VoxelMorph`-diff	4.20	0.049
`VoxelMorph`-huge	28.50	1.107
`CycleMorph`	41.90	0.281
`MIDIR`	4.05	1.627
`ViT-V-Net`	9.20	0.197
`PVT`	13.80	0.209
`CoTr`	17.10	0.372
`nnFormer`	6.35	0.105
`TransMorph-Bayes`	22.60	7.739
`TransMorph-diff`	7.35	0.099
`TransMorph-bspl`	10.50	1.739
`TransMorph`	14.40	0.329