RetinaNet pytorch implement from scratch 02--FPN

特征金字塔,用上采样+3x3卷积得到P3~P5

P6和P7通过C5得到

import torch.nn as nn

class FPN(nn.Module):
    def __init__(self, C3_size, C4_size, C5_size, feature_size=256):
        # feature_size 为feature map的channel数
        super(FPN, self).__init__()
        #P3~P6
        self.P5_1 = nn.Conv2d(C5_size, feature_size, kernel_size=1, stride=1, padding=0)
        # upsampling
        self.P5_upsampled = nn.Upsample(scale_factor=2, mode='nearest')
        self.P5_2 = nn.Conv2d(feature_size, feature_size, kernel_size=3, stride=1, padding=1)
        
        self.P4_1 = nn.Conv2d(C4_size, feature_size, kernel_size=1, stride=1, padding=0)
        self.P4_upsampled = nn.Upsample(scale_factor=2, mode='nearest')
        self.P4_2 = nn.Conv2d(feature_size, feature_size, kernel_size=3, stride=1, padding=1)

        self.P3_1 = nn.Conv2d(C3_size, feature_size, kernel_size=1, stride=1, padding=0)
        self.P3_2 = nn.Conv2d(feature_size, feature_size, kernel_size=3, stride=1, padding=1)
        
        # "P6 is obtained via a 3x3 stride-2 conv on C5"
        # if p6 use the P5 ,just change the argu C5_size to 256 !!!
        # self.P6 = nn.Conv2d(256, feature_size, kernel_size=3, stride=2, padding=1)
        self.P6 = nn.Conv2d(C5_size, feature_size, kernel_size=3, stride=2, padding=1)
        self.P7_1 = nn.ReLU()
        # =upsample
        self.P7_2 = nn.Conv2d(feature_size, feature_size, kernel_size=3, stride=2, padding=1)

    def forward(self, x):
        C3, C4, C5 = x
        P5_x = self.P5_1(C5)
        P5_upsampled_x = self.P5_upsampled(P5_x)
        P5_x = self.P5_2(P5_x)

        P4_x = self.P4_1(C4)
        P4_x = P5_upsampled_x + P4_x
        P4_upsampled_x = self.P4_upsampled(P4_x)
        P4_x = self.P4_2(P4_x)

        P3_x = self.P3_1(C3)
        P3_x = P3_x + P4_upsampled_x
        P3_x = self.P3_2(P3_x)

        P6_x = self.P6(C5)
        #P6_x = self.P6(P5_x)

        P7_x = self.P7_1(P6_x)
        P7_x = self.P7_2(P7_x)

        return [P3_x, P4_x, P5_x, P6_x, P7_x]