Skip to main content

Table 3 F1 of models with the proposed optimization strategies based on Table 2 (except GapNet)

From: Multi-labelled proteins recognition for high-throughput microscopy images using deep convolutional neural networks

Model

TA

TE

TA + TE

ResNet50 (fold 0)

0.774

0.772

0.774

ResNet50 (fold 1)

0.769

0.771

0.773

ResNet50 (fold 2)

0.773

0.769

0.775

ResNet50 (fold 3)

0.763

0.760

0.765

ResNet50 (fold 4)

0.765

0.763

0.767

InceptionV4 (single fold)

0.772

0.770

0.772

SENet (single fold)

0.786

0.783

0.789

Dense50 (random fold 1)

0.762

0.760

0.765

Dense50 (random fold 2)

0.765

0.763

0.766

Dense50 (random fold 3)

0.771

0.769

0.774

ResNet101 (single fold)

0.782

0.784

0.785

GapNet-PL

0.765

0.765

0.765

  1. The bolds of “0.786, 0.784 and 0.789” mean that SENet, ResNet101 and SENet can reach up to the best macro F1 score for TA, TE and TA+TE, respectively