Table 2 Classification results for a ResNet-50 fine-tuned on bubbles and dirt.

From: Galar - a large multi-label video capsule endoscopy dataset

Label

F1

AUROC

Accuracy

bubbles

0.89

0.87

0.82

dirt

0.88

0.94

0.88

macro average

0.88

0.90

0.85

micro average

0.89

—

0.85

  1. The metrics were computed individually for each label, and both macro- and micro-averaged scores are calculated across all labels. The outcomes are averaged across the 5 cross-validation folds.
Back to article page