Skip to main content

Table 1 Comparison of the model performance (F1) between original and modified ones

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

Model

Original

Customized

With hard sampler

ResNet50 (fold 0)

0.737

0.750

0.761

ResNet50 (fold 1)

0.729

0.734

0.752

ResNet50 (fold 2)

0.735

0.746

0.758

ResNet50 (fold 3)

0.731

0.741

0.749

ResNet50 (fold 4)

0.726

0.730

0.755

InceptionV4 (single fold)

0.736

0.749

0.756

SENet (single fold)

0.745

0.736

0.771

Dense50 (random fold 1)

0.732

0.738

0.741

Dense50 (random fold 2)

0.729

0.735

0.748

Dense50 (random fold 3)

0.732

0.741

0.754

ResNet101 (single fold)

0.742

0.758

0.760

  1. The bold of “0.771” means that SENet with hard sampler performs best for the model performance (F1)