Skip to main content
. 2017 Aug 11;17:181. doi: 10.1186/s12862-017-1014-z

Table 1.

GoogleNet architecture modified with Batch Normalization

Type Patch size / Stride Output size Depth Params Ops
Convolution 7 ×7/2 112 ×112×64 1 2.7K 34M
Max pool 3 ×3/2 56 ×56×64 0
Batch norm 56 ×56×64 0
LRN 56 ×56×64 0
Convolution 3 ×3/1 56 ×56×192 2 112K 360M
Max pool 3 ×3/2 28 ×28×192 0
Batch norm 28 ×28×192 0
LRN 28 ×28×192 0
Inception (3a) 28 ×28×256 2 159K 128M
Inception (3b) 28 ×28×480 2 380K 304M
Max pool 3 ×3/2 14 ×14×480 0
Batch norm 14 ×14×480 0
Inception (4a) 14 ×14×512 2 364K 73M
Inception (4b) 14 ×14×512 2 437K 88M
inception (4c) 14 ×14×512 2 463K 100M
Inception (4d) 14 ×14×528 2 580K 119M
Inception (4e) 14 ×14×832 2 840K 170M
Max pool 3 ×3/2 7 ×7×832 0
Batch norm 7 ×7×832 0
Inception (5a) 7 ×7×832 2 1072K 54M
Inception (5b) 7 ×7×1024 2 1388K 71M
Avg pool 7 ×7/1 1 ×1×1024 0
Batch norm 1 ×1×1024 0
Linear 1 ×1×10000 1 1000K 1M
Softmax 1 ×1×10000 0