MXNet实现Alexnet深度学习框架

Alexnet特点总结：

Alexnet会将输入的图像缩放到256*256，再从中随机截取出224*224的区域作为输入。

输入图像3*224*224 经过5次卷积和2次最大池化变为256*12*12。

1. Alexnet中的非线性激活都采用relu
2. Alexnet中的最大池化均为大小3，步长2
3. Alexnet卷积中使用了外衬

再经过1次最大池化，变成256*5*5，摊平后变成6400，经过3个全连接层，变成1000个数字，再进过softmax对1000个类别进行分类。

使用mxnet实现模型的框架代码如下：

import logging
import math
import random
import mxnet as mx
import numpy as np

#Alexnet
data = mx.sym.Variable('data')
#3 * 224 * 224 => 96 * 54 * 54
conv1 = mx.sym.Convolution(name='conv1',data=data,kernel=(11,11),stride=(4,4),num_filter=96)
relu1 = mx.sym.Activation(data=conv1,act_type='relu')
#96 * 54 * 54 => 96 * 26 * 26
pool1 = mx.sym.Pooling(data=relu1,pool_type='max',kernel=(3,3),stride=(2,2))

#96 * 26 * 26 =>  256 * 26 * 26
conv2 = mx.sym.Convolution(name='conv2',data=pool1,kernel=(5,5),stride=(2,2),num_filter=256)
relu2 = mx.sym.Activation(data=conv2,act_type='relu')
#256 * 26 * 26 => 256 * 12 * 12
pool2 = mx.sym.Pooling(data=relu2,pool_type='max',kernel=(3,3),stride=(2,2))

# 256 * 12 * 12 => 384 * 12 * 12
conv3 = mx.sym.Convolution(name='conv3',data=pool2,kernel=(3,3),pad=(1,1),num_filter=384)
relu3 = mx.sym.Activation(data=conv3,act_type='relu')

# 384 * 12 * 12 => 384 * 12 * 12
conv4 = mx.sym.Convolution(name='conv4',data=relu3,kernel=(3,3),pad=(1,1),num_filter=384)
relu4 = mx.sym.Activation(data=conv4,act_type='relu')

# 384 * 12 * 12 => 256 * 12 * 12
conv5 = mx.sym.Convolution(name='conv5',data=relu4,kernel=(3,3),pad=(1,1),num_filter=256)
relu5 = mx.sym.Activation(data=conv5,act_type='relu')
# 256 * 12 * 12 => 256* 5 * 5
pool3 = mx.sym.Pooling(data=relu5,kernel=(3,3),stride=(2,2),pool_type='max')

# 256 * 5 * 5 => 6400
flatten = mx.sym.Flatten(data=pool3)

#6400 => 4096
fc1 = mx.sym.FullyConnected(name='fc1',data=flatten,num_hidden=4096)
relu6 = mx.sym.Activation(data=fc1,act_type='relu')
dropout1 = mx.sym.Dropout(data=relu6,p=0.5)

#4096 => 4096
fc2 = mx.sym.FullyConnected(name='fc2',data=dropout1,num_hidden=4096)
relu7 = mx.sym.Activation(data=fc2,act_type='relu')
dropout2 = mx.sym.Dropout(data=relu7,p=0.5)

#4096 => 1000
fc3 = mx.sym.FullyConnected(name='fc3',data=dropout2,num_hidden=1000)
softmax = mx.sym.SoftmaxOutput(data=fc3,name='softmax')

shape = {'data' : (32,3,224,224)}
mx.viz.print_summary(symbol=softmax,shape=shape)