模型压缩和 网络参数量计算的网页

1.定点化

2.剪枝

剪channel,剪去绝对值较小的,即权重较小

剪一点fintune一次

https://arxiv.org/pdf/1611.06440v1.pdf

fpn的方式,fpga没办法用

fpga中1*1的卷积其实也是3*3,填充了0

http://dgschwend.github.io/netscope/#/editor

必须改成input_shape,即你必须告诉他输入的图片大小

name: "vgg_1/8"
input:"data"
input_shape{
dim:1
dim:3
dim:384
dim:768
}
layer {
name: "conv1_1"
type: "Convolution"
bottom: "data"
top: "conv1_1"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: 8
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.0
}
}
}

这个脚本识别Convolution,但像wk的ConvolutionData层就识别不了,不无法计算参数量的多少了

3.让模型更加过拟合,减小weight_decay