1 前言

FPN网络是一种常见的特征融合模块，在很多模型中都有运用，今天我们就结合代码和论文详细的搞清楚它到底是怎么一回事。

2 原理

原理直接看这一张图就可以了，很直观主要就是把对不同层的特征进行融合，重点还是在于代码的理解。

3 代码详解

import torch.nn as nn
import torch
import math
import torch.utils.model_zoo as model_zoo
from torchvision.ops import nms
# from retinanet.utils import BasicBlock, Bottleneck, BBoxTransform, ClipBoxes
# from retinanet.anchors import Anchors
# from retinanet import losses

class PyramidFeatures(nn.Module):
    def __init__(self, C3_size, C4_size, C5_size, feature_size=256):
        super(PyramidFeatures, self).__init__()

        # upsample C5 to get P5 from the FPN paper
        self.P5_1 = nn.Conv2d(C5_size, feature_size, kernel_size=1, stride=1, padding=0)
        self.P5_upsampled = nn.Upsample(scale_factor=2, mode='nearest')
        #将C5的特征图尺寸放大2倍用于跟C4相加
        self.P5_2 = nn.Conv2d(feature_size, feature_size, kernel_size=3, stride=1, padding=1)

        # add P5 elementwise to C4
        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)

        # add P4 elementwise to C3
        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"
        self.P6 = nn.Conv2d(C5_size, feature_size, kernel_size=3, stride=2, padding=1)

        # "P7 is computed by applying ReLU followed by a 3x3 stride-2 conv on P6"
        self.P7_1 = nn.ReLU()
        self.P7_2 = nn.Conv2d(feature_size, feature_size, kernel_size=3, stride=2, padding=1)

    def forward(self, inputs):
        #注意理解这里的inputs，其其表示的是一个列表
        C3, C4, C5 = inputs

        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)

        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]


if __name__ == '__main__':
    model = PyramidFeatures(32, 64, 96)
    print(model)
    ##这里假设输入是三层不同尺寸的特征图，输入的形状是[batch_size, 256, height, width]
    input = [torch.randn(1, 32, 640, 640), torch.randn(1, 64, 320, 320), torch.randn(1, 96, 160, 160)]
    out = model(input)
    print(out)