Lenet-5计算推理过程

LeNet-5的维度计算

C1-layer
(INPUT 32X32 ,conv kernel 5X5,Types of conv kernel 6)(Number of neurons 28X28X6)（Trainable parameters ）
#Note:2832-5+1

S2-layer
(INPUT 28X28X6,Sampling area 2X2,types of sampling 6)(feature map 14X14)(Number of neurons 14X14X6)
#Note:“Sampling area 2X2” means shrinking half, types of sampling is the same with z dimension6，
#Note:feature map 14=28/2

C2-layer
(INPUT 14X14X6,conv kernel 5X5,types of conv kernel 16)==(feature map 10X10)
#Note:“feature map”:10=14-5+1

S4-layer
(INPUT 10X10,Sampling are 2X2)(feature map 5X5)(Number of neurons 5X5X16)
#Note: 5=10/2，55

C5-layer (INPUT 5X5X16, conv kernel 5X5, Types of conv kernel 120)(Feature map 1X1)
#Note: 15-5+1

F6-layer
(INPUT 120, Number of neurons 84)==(Trainable parameters 10164)
#Note: 10164=84X(120+1) , 1 is a bias

Ouput-layer
(INPUT 10164 , Number of neurons 10)==(OUTPUT is a number from 0 to 9)