Skip to main content
. 2023 Jun 25;40(3):482–491. [Article in Chinese] doi: 10.7507/1001-5515.202302050

表 1. Performance comparison of different pre-training methods on the MVI dataset.

不同预训练方法在MVI数据集上的性能比较

算法 ACC AUC REC PRE F1
random (66.03 ± 3.39)% (68.56 ± 0.88)% (66.97 ± 6.24% (64.21 ± 4.66)% (65.15 ± 2.99)%
transfer (70.51 ± 1.11)% (72.42 ± 1.02)% (63.22 ± 3.87)% (76.27 ± 1.94)% (68.81 ± 1.99)%
SimCLR (73.40 ± 1.36)% (77.28 ± 2.26)% (81.91 ± 0.15)% (72.04 ± 1.51)% (76.61 ± 0.91)%
BYOL (72.44 ± 0.65)% (74.24 ± 2.91)% (79.95 ± 7.61)% (72.45 ± 2.45)% (75.77 ± 4.27)%
SimSiam (70.94 ± 2.43)% (70.43 ± 2.42)% (81.68 ± 4.48)% (69.27 ± 4.78)% (74.72 ± 2.93)%
MICLe (73.50 ± 2.25)% (81.13 ± 0.31)% (81.67 ± 5.88)% (64.57 ± 2.82)% (71.97 ± 1.09)%
Self-Trans (71.36 ± 0.74)% (81.04 ± 2.31)% (86.91 ± 1.68)% (66.31 ± 0.79)% (74.87 ± 0.84)%
SMR (72.86 ± 1.62)% (82.87 ± 1.83)% (86.98 ± 1.32)% (63.97 ± 2.12)% (73.70 ± 1.08)%
本文算法 (74.79 ± 0.74)% (83.60 ± 1.83)% (87.03 ± 4.25)% (71.48 ± 1.10)% (78.37 ± 1.94)%