毕业设计-基于深度学习的交通路面障碍物目标检测系统 YOLO python 卷积神经网络 人工智能

?

import torch
import torch.nn as nn

class FPN(nn.Module):
    def __init__(self, in_channels_list, out_channels):
        super(FPN, self).__init__()

        self.in_channels_list = in_channels_list
        self.num_layers = len(in_channels_list)
        self.out_channels = out_channels

        # Lateral convolutions
        self.lateral_convs = nn.ModuleList()
        for i in range(self.num_layers):
            conv = nn.Conv2d(in_channels_list[i], out_channels, kernel_size=1)
            self.lateral_convs.append(conv)

        # Top-down pathway convolutions
        self.topdown_convs = nn.ModuleList()
        for i in range(self.num_layers - 1, 0, -1):
            conv = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
            self.topdown_convs.append(conv)

        # Smooth convolution
        self.smooth_conv = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)

    def forward(self, inputs):
        # Lateral connections
        laterals = [lateral_conv(inputs[i]) for i, lateral_conv in enumerate(self.lateral_convs)]

        # Top-down pathway
        for i in range(self.num_layers - 1, 0, -1):
            _, _, H, W = laterals[i - 1].size()
            topdown = nn.functional.interpolate(laterals[i], size=(H, W), mode='nearest')
            laterals[i - 1] += topdown
            laterals[i - 1] = self.topdown_convs[i - 1](laterals[i - 1])

        # Smooth convolution
        outputs = [self.smooth_conv(laterals[i]) for i in range(self.num_layers)]

        return outputs

# Example usage
in_channels_list = [256, 512, 1024, 2048]  # Example input channels for each FPN level
out_channels = 256  # Desired output channels for FPN
fpn = FPN(in_channels_list, out_channels)

# Assuming inputs is a list of feature maps from different network levels
inputs = [torch.randn(1, c, h, w) for c, h, w in zip(in_channels_list, [64, 32, 16, 8])]
outputs = fpn(inputs)

# 'outputs' will be a list of feature maps from FPN, with the same number of levels as 'inputs'

?

?

import torch
from torchvision.models import yolov5

# 加载预训练的YOLOv5模型
model = yolov5s(pretrained=True)

# 设置模型为评估模式
model.eval()

# 定义类别标签
class_labels = ['obstacle', 'car', 'pedestrian', 'bicycle', 'motorcycle']

# 加载测试图像
image = torch.randn(1, 3, 416, 416)  # 示例输入图像，尺寸为(1, 3, 416, 416)

# 运行图像通过模型进行推理
with torch.no_grad():
    outputs = model(image)