Skip to main content
. 2023 Sep 7;23(18):7724. doi: 10.3390/s23187724

Table 2.

Comparative outline of Hyperparameter Settings with proposed model where (-) denotes the null value.

Method Trainable Parameter Learning Rate Momentum Optimizer Batch Size Loss Threshold
FCN8 [46] 134,270,278 1 × 10−4 - SGD 8 Cross-entropy -
HRNet [47] 9,524,036 1 × 10−4 - Adam 8 Dice loss -
PSPNet [48] 48,631,850 1 × 10−2 - SGD 8 Cross-entropy -
DeepLabv3+ [49] ResNet50: 39,756,962 1 × 10−2 - SGD 8 Cross-entropy -
ResNet101: 58,749,090 1 × 10−3 - SGD 8 Cross-entropy -
UNet 14,326,275 1 × 10−5 0.9 Nadam 16 BCE 0.5
Attention-UNet 8,135,745 1 × 10−5 0.9 Nadam 16 BCE 0.5
ResUNet 1,048,953 1 × 10−5 0.9 Nadam 16 BCE 0.5
ResUNet++ 16,228,001 1 × 10−5 0.9 Nadam 16 BCE 0.5
IRv2-Net
(Proposed)
28,864,481 1 × 10−5 0.9 Nadam 16 BCE 0.5