Skip to main content
. 2021 Nov 11;21(22):7498. doi: 10.3390/s21227498

Table 7.

Comparison of facial landmark feature extraction methods.

Method Accuracy (%)
ResNet-18 + Att 65.3396
ResNet-18 + Att + Landmark Image 63.0085
ResNet-18 + Att + Landmark Image + Norm 64.0012
ResNet-18 + Att + Landmark Image + Blur 66.1854
ResNet-18 + Att + Landmark Image + Norm + Blur 66.8409

Att: spatial and temporal attention modules, Norm: min–max normalization, Blur: Gaussian blurring.