源代碼

• 通過截取 vgg19 feature map 不同 size 特征段，獲取多種 size 的 feature map
• 具體為 5 種 feature map size: [1, 1/2, 1/4, 1/8, 1/8]

import torch.nn as nn
from torchvision import models


class VGG19(nn.Module):
    def __init__(self):
        super(VGG19, self).__init__()

        # 查看 vgg19 結構，得知 features 為 nn.Sequential，可用 idx 獲取特定 layer 參數
        features = models.vgg19(pretrained=True).features
        self.to_relu_1_1 = nn.Sequential()
        self.to_relu_2_1 = nn.Sequential()
        self.to_relu_3_1 = nn.Sequential()
        self.to_relu_4_1 = nn.Sequential()
        self.to_relu_4_2 = nn.Sequential()

        # 截取 vgg19 不同 feature map size 的特征段，參照下方源代碼
        for x in range(2):
            self.to_relu_1_1.add_module(str(x), features[x])   # 1/1
        for x in range(2, 7):
            self.to_relu_2_1.add_module(str(x), features[x])   # 1/2
        for x in range(7, 12):
            self.to_relu_3_1.add_module(str(x), features[x])   # 1/4
        for x in range(12, 21):
            self.to_relu_4_1.add_module(str(x), features[x])   # 1/8
        for x in range(21, 25):
            self.to_relu_4_2.add_module(str(x), features[x])   # 1/8

        for param in self.parameters():
            param.requires_grad = False

    def forward(self, x):
        h = self.to_relu_1_1(x)
        h_relu_1_1 = h
        h = self.to_relu_2_1(h)
        h_relu_2_1 = h
        h = self.to_relu_3_1(h)
        h_relu_3_1 = h
        h = self.to_relu_4_1(h)
        h_relu_4_1 = h
        h = self.to_relu_4_2(h)
        h_relu_4_2 = h
        out = (h_relu_1_1, h_relu_2_1, h_relu_3_1, h_relu_4_1, h_relu_4_2)
        return out   # 返回 [1, 1/2, 1/4, 1/8, 1/8] 5 種 size 的中間特征

用 print 直觀查看 vgg19 模型原始結構

from torchvision import models

vgg = models.vgg19(pretrained=False)
print(vgg)

卷積核輸出 feature map size

VGG(

  # features = models.vgg19(pretrained=True).features
  # 此項為 獲取的 pretrained vgg19 參數；如下注釋顯示 self.relu 使用的參數塊
  (features): Sequential(

     # self.to_relu_1_1  -> 1/1
    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (1): ReLU(inplace=True)

    # self.to_relu_2_1  -> 1/2
    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (3): ReLU(inplace=True)
    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (6): ReLU(inplace=True)

    # self.to_relu_3_1  -> 1/4
    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (8): ReLU(inplace=True)
    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (11): ReLU(inplace=True)

    # self.to_relu_4_1  -> 1/8
    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (13): ReLU(inplace=True)
    (14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (15): ReLU(inplace=True)
    (16): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (17): ReLU(inplace=True)
    (18): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (19): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (20): ReLU(inplace=True)

    # self.to_relu_4_2  -> 1/8
    (21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (22): ReLU(inplace=True)
    (23): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (24): ReLU(inplace=True)

    # 之后參數，沒有使用
    (25): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (26): ReLU(inplace=True)
    (27): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
    (28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (29): ReLU(inplace=True)
    (30): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (31): ReLU(inplace=True)
    (32): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (33): ReLU(inplace=True)
    (34): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    (35): ReLU(inplace=True)
    (36): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
  (classifier): Sequential(
    (0): Linear(in_features=25088, out_features=4096, bias=True)
    (1): ReLU(inplace=True)
    (2): Dropout(p=0.5, inplace=False)
    (3): Linear(in_features=4096, out_features=4096, bias=True)
    (4): ReLU(inplace=True)
    (5): Dropout(p=0.5, inplace=False)
    (6): Linear(in_features=4096, out_features=1000, bias=True)
  )
)