Table 3.

The network structure of GoogLeNet.

Type	Depth	Stride	Size(output)	#1 × 1	#3 × 3 Reduce	#3 × 3	#5 × 5 Reduce	#5 × 5	PP	Params
Conv	1	7 × 7/2	112 × 112 × 64							2.7 K
Max pool	0	3 × 3/2	56 × 56 × 64
Conv	2	3 × 3/1	56 × 56 × 192		64	192				112 K
Max pool	0	3 × 3/2	56 × 56 × 192
Inception (3a)	2		28 × 28 × 256	64	96	128	16	32	32	159 K
Inception (3b)	2		28 × 28 × 480	128	128	192	32	96	64	380 K
Max pool	0	3 × 3/2	14 × 14 × 480
Inception (4a)	2		14 × 14 × 512	192	96	208	16	48	64	364 K
Inception (4b)	2		14 × 14 × 512	160	112	224	24	64	64	437 K
Inception (4c)	2		14 × 14 × 512	128	128	256	24	64	64	463 K
Inception (4d)	2		14 × 14 × 528	112	144	288	32	64	64	580 K
Inception (4e)	2		14 × 14 × 832	256	160	320	32	128	128	840 K
Max pool	0	3 × 3/2	7 × 7 × 832
Inception (5a)	2		7 × 7 × 832	256	160	320	32	128	128	1072 K
Inception (5b)	2		7 × 7 × 1024	384	192	384	48	128	128	1388 K
Avg pool	0	7 × 7/1	1×1 × 1024
Dropout (40%)	0		1 × 1 × 1024
Linear	1		1 × 1 × 1000							1000 K
Softmax	0		1 × 1 × 1000