pytorch在fintune時(shí)將sequential中的層輸出方法,以vgg為例

更新時(shí)間：2019年08月20日 08:40:24 作者：BinWang-cvlab

今天小編就為大家分享一篇pytorch在fintune時(shí)將sequential中的層輸出方法,以vgg為例，具有很好的參考價(jià)值，希望對(duì)大家有所幫助。一起跟隨小編過來看看吧

有時(shí)候我們?cè)趂intune時(shí)發(fā)現(xiàn)pytorch把許多層都集合在一個(gè)sequential里，但是我們希望能把中間層的結(jié)果引出來做下一步操作，于是我自己琢磨了一個(gè)方法，以vgg為例，有點(diǎn)僵硬哈！

首先pytorch自帶的vgg16模型的網(wǎng)絡(luò)結(jié)構(gòu)如下：

VGG(
 (features): Sequential(
 (0): Conv2d (3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (1): ReLU(inplace)
 (2): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (3): ReLU(inplace)
 (4): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1))
 (5): Conv2d (64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (6): ReLU(inplace)
 (7): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (8): ReLU(inplace)
 (9): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1))
 (10): Conv2d (128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (11): ReLU(inplace)
 (12): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (13): ReLU(inplace)
 (14): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (15): ReLU(inplace)
 (16): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1))
 (17): Conv2d (256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (18): ReLU(inplace)
 (19): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (20): ReLU(inplace)
 (21): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (22): ReLU(inplace)
 (23): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1))
 (24): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (25): ReLU(inplace)
 (26): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (27): ReLU(inplace)
 (28): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
 (29): ReLU(inplace)
 (30): MaxPool2d(kernel_size=(2, 2), stride=(2, 2), dilation=(1, 1))
 )
 (classifier): Sequential(
 (0): Linear(in_features=25088, out_features=4096)
 (1): ReLU(inplace)
 (2): Dropout(p=0.5)
 (3): Linear(in_features=4096, out_features=4096)
 (4): ReLU(inplace)
 (5): Dropout(p=0.5)
 (6): Linear(in_features=4096, out_features=1000)
 )
)

我們需要fintune vgg16的features部分，并且我希望把3，8， 15， 22， 29這五個(gè)作為輸出進(jìn)一步操作。我的想法是自己寫一個(gè)vgg網(wǎng)絡(luò)，這個(gè)網(wǎng)絡(luò)參數(shù)與pytorch的網(wǎng)絡(luò)一致但是保證我們需要的層輸出在sequential外。于是我寫的網(wǎng)絡(luò)如下：

class our_vgg(nn.Module):
 def __init__(self):
  super(our_vgg, self).__init__()
  self.conv1 = nn.Sequential(
   # conv1
   nn.Conv2d(3, 64, 3, padding=35),
   nn.ReLU(inplace=True),
   nn.Conv2d(64, 64, 3, padding=1),
   nn.ReLU(inplace=True),

  )
  self.conv2 = nn.Sequential(
   # conv2
   nn.MaxPool2d(2, stride=2, ceil_mode=True), # 1/2
   nn.Conv2d(64, 128, 3, padding=1),
   nn.ReLU(inplace=True),
   nn.Conv2d(128, 128, 3, padding=1),
   nn.ReLU(inplace=True),

  )
  self.conv3 = nn.Sequential(
   # conv3
   nn.MaxPool2d(2, stride=2, ceil_mode=True), # 1/4
   nn.Conv2d(128, 256, 3, padding=1),
   nn.ReLU(inplace=True),
   nn.Conv2d(256, 256, 3, padding=1),
   nn.ReLU(inplace=True),
   nn.Conv2d(256, 256, 3, padding=1),
   nn.ReLU(inplace=True),

  )
  self.conv4 = nn.Sequential(
   # conv4
   nn.MaxPool2d(2, stride=2, ceil_mode=True), # 1/8
   nn.Conv2d(256, 512, 3, padding=1),
   nn.ReLU(inplace=True),
   nn.Conv2d(512, 512, 3, padding=1),
   nn.ReLU(inplace=True),
   nn.Conv2d(512, 512, 3, padding=1),
   nn.ReLU(inplace=True),

  )
  self.conv5 = nn.Sequential(
   # conv5
   nn.MaxPool2d(2, stride=2, ceil_mode=True), # 1/16
   nn.Conv2d(512, 512, 3, padding=1),
   nn.ReLU(inplace=True),
   nn.Conv2d(512, 512, 3, padding=1),
   nn.ReLU(inplace=True),
   nn.Conv2d(512, 512, 3, padding=1),
   nn.ReLU(inplace=True),
  )


 def forward(self, x):

  conv1 = self.conv1(x)
  conv2 = self.conv2(conv1)
  conv3 = self.conv3(conv2)
  conv4 = self.conv4(conv3)
  conv5 = self.conv5(conv4)

  return conv5

接著就是copy weights了：

def convert_vgg(vgg16):#vgg16是pytorch自帶的
 net = our_vgg()# 我寫的vgg

 vgg_items = net.state_dict().items()
 vgg16_items = vgg16.items()

 pretrain_model = {}
 j = 0
 for k, v in net.state_dict().iteritems():#按順序依次填入
  v = vgg16_items[j][1]
  k = vgg_items[j][0]
  pretrain_model[k] = v
  j += 1
 return pretrain_model


## net是我們最后使用的網(wǎng)絡(luò)，也是我們想要放置weights的網(wǎng)絡(luò)
net = net()

print ('load the weight from vgg')
pretrained_dict = torch.load('vgg16.pth')
pretrained_dict = convert_vgg(pretrained_dict)
model_dict = net.state_dict()
# 1. 把不屬于我們需要的層剔除
pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
# 2. 把參數(shù)存入已經(jīng)存在的model_dict
model_dict.update(pretrained_dict) 
# 3. 加載更新后的model_dict
net.load_state_dict(model_dict)
print ('copy the weight sucessfully')

這樣我就基本達(dá)成目標(biāo)了，注意net也就是我們要使用的網(wǎng)絡(luò)fintune部分需要和our_vgg一致。

以上這篇pytorch在fintune時(shí)將sequential中的層輸出方法,以vgg為例就是小編分享給大家的全部?jī)?nèi)容了，希望能給大家一個(gè)參考，也希望大家多多支持腳本之家。

您可能感興趣的文章: