获取Pytorch中间某一层权重或者特征

问题:训练好的网络模型想知道中间某一层的权重或者看看中间某一层的特征,如何处理呢?

1.获取某一层权重,并保存到excel中;

以resnet18为例说明:

import torch
import pandas as pd
import numpy as np
import torchvision.models as models

resnet18 = models.resnet18(pretrained=True)

parm={}
for name,parameters in resnet18.named_parameters():
    print(name,':',parameters.size())
    parm[name]=parameters.detach().numpy()

上述代码将每个模块参数存入parm字典中,parameters.detach().numpy()将tensor类型变量转换成numpy array形式,方便后续存储到表格中.输出为:

conv1.weight : torch.Size([64, 3, 7, 7])
bn1.weight : torch.Size([64])
bn1.bias : torch.Size([64])
layer1.0.conv1.weight : torch.Size([64, 64, 3, 3])
layer1.0.bn1.weight : torch.Size([64])
layer1.0.bn1.bias : torch.Size([64])
layer1.0.conv2.weight : torch.Size([64, 64, 3, 3])
layer1.0.bn2.weight : torch.Size([64])
layer1.0.bn2.bias : torch.Size([64])
layer1.1.conv1.weight : torch.Size([64, 64, 3, 3])
layer1.1.bn1.weight : torch.Size([64])
layer1.1.bn1.bias : torch.Size([64])
layer1.1.conv2.weight : torch.Size([64, 64, 3, 3])
layer1.1.bn2.weight : torch.Size([64])
layer1.1.bn2.bias : torch.Size([64])
layer2.0.conv1.weight : torch.Size([128, 64, 3, 3])
layer2.0.bn1.weight : torch.Size([128])
layer2.0.bn1.bias : torch.Size([128])
layer2.0.conv2.weight : torch.Size([128, 128, 3, 3])
layer2.0.bn2.weight : torch.Size([128])
layer2.0.bn2.bias : torch.Size([128])
layer2.0.downsample.0.weight : torch.Size([128, 64, 1, 1])
layer2.0.downsample.1.weight : torch.Size([128])
layer2.0.downsample.1.bias : torch.Size([128])
layer2.1.conv1.weight : torch.Size([128, 128, 3, 3])
layer2.1.bn1.weight : torch.Size([128])
layer2.1.bn1.bias : torch.Size([128])
layer2.1.conv2.weight : torch.Size([128, 128, 3, 3])
layer2.1.bn2.weight : torch.Size([128])
layer2.1.bn2.bias : torch.Size([128])
layer3.0.conv1.weight : torch.Size([256, 128, 3, 3])
layer3.0.bn1.weight : torch.Size([256])
layer3.0.bn1.bias : torch.Size([256])
layer3.0.conv2.weight : torch.Size([256, 256, 3, 3])
layer3.0.bn2.weight : torch.Size([256])
layer3.0.bn2.bias : torch.Size([256])
layer3.0.downsample.0.weight : torch.Size([256, 128, 1, 1])
layer3.0.downsample.1.weight : torch.Size([256])
layer3.0.downsample.1.bias : torch.Size([256])
layer3.1.conv1.weight : torch.Size([256, 256, 3, 3])
layer3.1.bn1.weight : torch.Size([256])
layer3.1.bn1.bias : torch.Size([256])
layer3.1.conv2.weight : torch.Size([256, 256, 3, 3])
layer3.1.bn2.weight : torch.Size([256])
layer3.1.bn2.bias : torch.Size([256])
layer4.0.conv1.weight : torch.Size([512, 256, 3, 3])
layer4.0.bn1.weight : torch.Size([512])
layer4.0.bn1.bias : torch.Size([512])
layer4.0.conv2.weight : torch.Size([512, 512, 3, 3])
layer4.0.bn2.weight : torch.Size([512])
layer4.0.bn2.bias : torch.Size([512])
layer4.0.downsample.0.weight : torch.Size([512, 256, 1, 1])
layer4.0.downsample.1.weight : torch.Size([512])
layer4.0.downsample.1.bias : torch.Size([512])
layer4.1.conv1.weight : torch.Size([512, 512, 3, 3])
layer4.1.bn1.weight : torch.Size([512])
layer4.1.bn1.bias : torch.Size([512])
layer4.1.conv2.weight : torch.Size([512, 512, 3, 3])
layer4.1.bn2.weight : torch.Size([512])
layer4.1.bn2.bias : torch.Size([512])
fc.weight : torch.Size([1000, 512])
fc.bias : torch.Size([1000])

parm['layer1.0.conv1.weight'][0,0,:,:]

输出为:

array([[ 0.05759342, -0.09511436, -0.02027232],
       [-0.07455588, -0.799308  , -0.21283598],
       [ 0.06557069, -0.09653367, -0.01211061]], dtype=float32)

利用如下函数将某一层的所有参数保存到表格中,数据维持卷积核特征大小,如3*3的卷积保存后还是3x3的.

def parm_to_excel(excel_name,key_name,parm):
    with pd.ExcelWriter(excel_name) as writer:
        [output_num,input_num,filter_size,_]=parm[key_name].size()
        for i in range(output_num):
            for j in range(input_num):
                data=pd.DataFrame(parm[key_name][i,j,:,:].detach().numpy())
                #print(data)
                data.to_excel(writer,index=False,header=True,startrow=i*(filter_size+1),startcol=j*filter_size)

由于权重矩阵中有很多的值非常小,取出固定大小的值,并将全部权重写入excel

counter=1
with pd.ExcelWriter('test1.xlsx') as writer:
    for key in parm_resnet50.keys():
        data=parm_resnet50[key].reshape(-1,1)
        data=data[data>0.001]
        
        data=pd.DataFrame(data,columns=[key])
        data.to_excel(writer,index=False,startcol=counter)
        counter+=1

2.获取中间某一层的特性

重写一个函数,将需要输出的层输出即可.

def resnet_cifar(net,input_data):
    x = net.conv1(input_data)
    x = net.bn1(x)
    x = F.relu(x)
    x = net.layer1(x)
    x = net.layer2(x)
    x = net.layer3(x)
    x = net.layer4[0].conv1(x)  #这样就提取了layer4第一块的第一个卷积层的输出
    x=x.view(x.shape[0],-1)
    return x

model = models.resnet18()
x = resnet_cifar(model,input_data)

原文：https://blog.csdn.net/happyday_d/article/details/88974361