Table 3 Comparison between randomly initialized and ImageNet pretrained models.

From: A realistic fish-habitat dataset to evaluate algorithms for underwater visual analysis

 

Classification

Counting

Localization

Segmentation

Accuracy

MAE

MAE

GAME

mIoU

Random weights

0.65

1.30

1.22

1.30

0.49

Pretrained weights

0.99

0.38

0.21

1.22

0.93

  1. Classification results were evaluated on the FishClf dataset, counting and localization on the FishLoc dataset, and segmentation on the FishSeg dataset.
Back to article page