深度学习网络结构

imagenet-vgg-f.mat 网络结构：
conv1 conv size: 11 11 3 64%64是卷积核个数
relu1 relu
norm1 lrn
pool1 pool
conv2 conv size: 5 5 64 256%256同样也是卷积核个数，下面类似
relu2 relu
norm2 lrn
pool2 pool
conv3 conv size: 3 3 256 256
relu3 relu
conv4 conv size: 3 3 256 256
relu4 relu
conv5 conv size: 3 3 256 256
relu5 relu
pool5 pool
fc6 conv size: 6 6 256 4096
relu6 relu
fc7 conv size: 1 1 4096 4096
relu7 relu
fc8 conv size: 1 1 4096 1000
prob softmax
vggf-edgeboxes-boxsc-spatreg.mat
conv1 size: 11 11 3 64
relu1
norm1
pool1
conv2 size: 5 5 64 256
relu2
norm2
pool2
conv3 size: 3 3 256 256
relu3
conv4 size: 3 3 256 256
relu4
conv5 size: 3 3 256 256
relu5
fc6 size: 1 36 256 4096
relu6
drop6
fc7 size: 1 1 4096 4096
relu7
drop7
fc8C size: 1 1 4096 20
fc8R size: 1 1 4096 20
SPP
boostBox
softmaxDet
softmaxCls
timesCR
sum

mnist net

conv
bnorm
pool
conv
bnorm
pool
conv
bnorm
relu
conv
softmaxloss