Table 5 DL classification results: frames are aggregated over each video using majority voting.

From: Explainable automated recognition of emotional states from canine facial expressions: the case of positive anticipation and frustration

Backbone

Accuracy

Positive

Negative

Initial weights

Model

Precision

Recall

F1

Precision

Recall

F1

Supervised

ResNet50

0.81

0.91

0.70

0.79

0.75

0.93

0.83

ViT

0.82

0.80

0.87

0.83

0.85

0.77

0.81

DINO

ResNet50

0.81

0.92

0.70

0.79

0.75

0.94

0.83

ViT

0.89

0.94

0.84

0.89

0.85

0.95

0.90

  1. Best value is in bold.
Back to article page