[人工智能]caffe模型1X1卷积与FC互转

在caffe模型实际落地时，局限于硬件的问题，有时候训练好的网络结构需要进行微调，最近碰到NT98528输出不支持的问题，需要临时将1*1 conv改fc，先改prototxt，对应改下类型名称即可：

#caffe 卷积层
layer {
  name: "fc_1"
  type: "Convolution"
  bottom: "pool5"
  top: "fc_1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 1
  }
  convolution_param {
    num_output: 3
    kernel_size: 1
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}    



#caffe 全连接层
layer {
  name: "fc_1"
  type: "InnerProduct"
  bottom: "pool5"
  top: "fc_1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 1
  }
  inner_product_param {
    num_output: 3
    weight_filler {
      type: "msra"
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}    
    

1*1 conv和FC相互转化 代码如下：

# -*- coding: utf-8 -*-
#encoding: utf-8

import sys
import cv2
import math
import numpy as np
import os
from PIL import Image
import io

caffe_path = r'../../python'
sys.path.insert(0, caffe_path)
import caffe

params = ['fc_1', 'fc_2']
params_full_conv=params
 
def convert_full_conv(model_define, model_weight, model_define_fc, model_weight_fc):
    '''
    @breif: convert fc to conv
    @param: model_define, src_prototxt
    @param: model_weight, src_model
    @param: model_define_fc, dst_prototxt
    @param: model_weight_fc, dst_model
    '''
    net = caffe.Net(model_define, model_weight, caffe.TEST)
    fc_params = {pr: (net.params[pr][0].data, net.params[pr][1].data) for pr in params}
    net_fc = caffe.Net(model_define_fc, caffe.TEST)
    conv_params = {pr: (net_fc.params[pr][0].data, net_fc.params[pr][1].data) for pr in params_full_conv}

    for fc in params:
        print ('{} weights are {} dimensional and biases are {} dimensional'.format(fc, fc_params[fc][0].shape, fc_params[fc][1].shape))
    for conv in params_full_conv:
        print ('{} weights are {} dimensional and biases are {} dimensional'.format(conv, conv_params[conv][0].shape, conv_params[conv][1].shape))
        
    
    for pr, pr_conv in zip(params, params_full_conv):
        conv_params[pr_conv][0].flat = fc_params[pr][0].flat  # flat unrolls the arrays
        conv_params[pr_conv][1][...] = fc_params[pr][1]

    net_fc.save(model_weight_fc)
    print ('convert done!')
    return net_fc
 
if __name__ == '__main__':
 
    file  = 'lenet.prototxt'#原始的prototxt名称
    model = 'lenet.caffemodel'#原始的caffemodel名称
    conv_file = 'lenet_conv.prototxt' #修改后的prototxt
    conv_model = 'lenet_conv.caffemodel'#最终得到的结果

    convert_full_conv(file, model, conv_file, conv_model)

验证参数：

    file  = 'lenet.prototxt'#原始的prototxt名称
    model = 'lenet.caffemodel'#原始的caffemodel名称
    conv_file = 'lenet_conv.prototxt' #修改后的prototxt
    conv_model = 'lenet_conv.caffemodel'#最终得到的结果

    net = caffe.Net(file, model, caffe.TEST)
    fc_params = {pr: (net.params[pr][0].data, net.params[pr][1].data) for pr in params}
    net_fc = caffe.Net(conv_file, conv_model, caffe.TEST)
    conv_params = {pr: (net_fc.params[pr][0].data, net_fc.params[pr][1].data) for pr in params_full_conv}

    for pr, pr_conv in zip(params, params_full_conv):
        print(fc_params[pr][0].flat)
        print(conv_params[pr][0].flat)
        print(fc_params[pr][1])
        print(conv_params[pr][1])

底层硬件限制是真特么多，诸如卷积层通道不能超过2048，pool层尺寸不能大于6*6，pool输出不能超过16个等等，芯片厂商自己搞自己的，模型换平台每次都改，气死偶来。

参考文档：

https://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/net_surgery.ipynb

https://blog.csdn.net/andeyeluguo/article/details/79219709

关注博主即可阅读全文