Skip to main content
. 2022 Jul 5;8:e1031. doi: 10.7717/peerj-cs.1031

Table 2. Summary of deep architectures used in this work.

Architecture No. of
Conv layers		 No. of
FCC layers		 No. of training
parameters		 Minimum
image size		 Number of
extracted features		 Top 5 error
on ImageNet
DenseNet121 120 1 7 million 221×221 1,024 7.71%
InceptionV3 42 1 22 million 299 × 299 2,048 3.08%
VGG16 13 3 134 million 227 × 227 4,096 7.30%
MobileNet 53 3 3.4 million 224 × 224 1,024 -%