Table 5 LFAGCU model performance under various hyperparameter configurations.

From: Local feature acquisition and global context understanding network for very high-resolution land cover classification

 

LFAGCU

LFAGCU (M)

LFAGCU (S)

Into/Out GFL-channels

64/160

48/96

24/80

Attention-channels

144/240

96/144

64/96

GFL-dimension

288/480

192/288

128/192

Total params

2,664,120

1,199,064

771,928

Trainable params

2,664,120

1,199,064

771,928

Total mult-adds (M)

28.27

14.48

10.54

Forward/backward pass size (MB)

3.87

3.56

3.45

Params size (MB)

10.16

4.57

2.94

Estimated total size (MB)

14.60

8.71

6.97

  1. The upper part of the table presents data in the format of: 'Initial Layer/Final Layer'.
Back to article page