Skip to main content

Table 7 Weights size and FLOPs improvement by using our channel pruning scheme under different pruning ratios

From: Breast cancer histopathology image classification through assembling multiple compact CNNs

Method

Pruning Degree

Weights (M)

FLOP (M)

Our

Before Pruning

3.77

2920.3

 

Pruning Ratio 0.5

1.76

1861.8

 

Pruning Ratio 0.6

1.4

1663.1

 

Pruning Ratio 0.7

1.06

1477.8

 

Pruning Ratio 0.8

0.74

1305

 

Pruning Ratio 0.9

0.43

1133.6

 

Pruning Ratio 0.95

0.29

1063.8

Work[11]

N/A

0.55

47.4/188.5

Work[17]

N/A

13.5

8521

  1. The weights and FLOPs of work [11] and [17] are also included in the table. The work [11] has two types of networks with different input sizes: 32 ×32 and 64 ×64, and the corresponding FLOPs are 47.4 (M) and 188.5 (M), respectively