Table 2.

Mean absolute motion prediction errors without and with each conversion method (in mm, mean ± standard deviation) with the best results marked in bold.

	No Conversion	Vanilla GAN One-to-one	Vanilla GAN	MSE loss GAN	TAI-GAN
All frames	4.45 ± 0.64*	–	4.40 ± 0.49*	4.76 ± 0.48*	3.48 ± 0.45
EQ-1	6.18 ± 1.51*	5.33 ± 1.34	6.03 ± 1.06*	5.12 ± 0.72*	5.06 ± 0.78
EQ+1	5.12 ± 0.93*	4.72 ± 0.86	4.93 ± 0.80*	4.81 ± 0.46*	4.35 ± 0.87

^*P < 0.05 between the current class and TAI-GAN (paired-wise paired two-tailed t-test).