Skip to main content
Journal of Big Data

Table 3 Efficiency comparison of eight object localization methods

From: Automatic diagnosis of keratitis using object localization combined with cost-sensitive deep attention convolutional neural network

Methods

Size

Parameters(M)

Training time

Testing time

Faster R-CNN1

315.0 MB

4.1e + 07

1.41 h

0.078 s

Faster R-CNN2

460.3 MB

6.0e + 07

1.96 h

0.090 s

Cascade R-CNN1

527.1 MB

6.9e + 07

1.86 h

0.083 s

Cascade R-CNN2

672.4 MB

8.7e + 07

2.35 h

0.101 s

TridentNet

251.4 MB

3.3e + 07

3.89 h

0.068 s

RetinaNet1

276.9 MB

3.6e + 07

0.85 h

0.063 s

RetinaNet2

422.1 MB

5.5e + 07

1.27 h

0.071 s

SSD

187.2 MB

2.5e + 07

2.27 h

0.049 s

  1. Bold values represent the best performance among different methods in the same column. Training time indicates the running time of the method in the whole training process. Testing time indicates the average time that the method needs in testing every slit-lamp image. MB Mbyte
Back to article page