模式识别与图像处理课程实验二：基于UNet的目标检测网络_unet目标检测(1)

先自我介绍一下，小编浙江大学毕业，去过华为、字节跳动等大厂，目前阿里P7

深知大多数程序员，想要提升技能，往往是自己摸索成长，但自己不成体系的自学效果低效又漫长，而且极易碰到天花板技术停滞不前！

因此收集整理了一份《2024年最新大数据全套学习资料》，初衷也很简单，就是希望能够帮助到想自学提升又不知道该从何学起的朋友。

既有适合小白学习的零基础资料，也有适合3年以上经验的小伙伴深入学习提升的进阶课程，涵盖了95%以上大数据知识点，真正体系化！

由于文件比较多，这里只是将部分目录截图出来，全套包含大厂面经、学习笔记、源码讲义、实战项目、大纲路线、讲解视频，并且后续会持续更新

如果你需要这些资料，可以添加V获取：vip204888 （备注大数据）

正文

    self.relu = nn.ReLU(inplace=True)

    self.conv3 = nn.Conv2d(out_channels, out_channels, 3, padding=1)

    self.bn2 = nn.BatchNorm2d(out_channels)

def forward(self, x):
    # b, 10, 8, 1024
    batch_size = x.shape[0]
    out1_1 = self.conv1_1(nn.functional.pad(x, (1, 1, 1, 1)))
    out1_2 = self.conv1_2(nn.functional.pad(x, (1, 1, 0, 1)))#right interleaving padding
    #out1\_2 = self.conv1\_2(nn.functional.pad(x, (1, 1, 1, 0)))#author's interleaving pading in github
    out1_3 = self.conv1_3(nn.functional.pad(x, (0, 1, 1, 1)))#right interleaving padding
    #out1\_3 = self.conv1\_3(nn.functional.pad(x, (1, 0, 1, 1)))#author's interleaving pading in github
    out1_4 = self.conv1_4(nn.functional.pad(x, (0, 1, 0, 1)))#right interleaving padding
    #out1\_4 = self.conv1\_4(nn.functional.pad(x, (1, 0, 1, 0)))#author's interleaving pading in github

    out2_1 = self.conv2_1(nn.functional.pad(x, (1, 1, 1, 1)))
    out2_2 = self.conv2_2(nn.functional.pad(x, (1, 1, 0, 1)))#right interleaving padding
    #out2\_2 = self.conv2\_2(nn.functional.pad(x, (1, 1, 1, 0)))#author's interleaving pading in github
    out2_3 = self.conv2_3(nn.functional.pad(x, (0, 1, 1, 1)))#right interleaving padding
    #out2\_3 = self.conv2\_3(nn.functional.pad(x, (1, 0, 1, 1)))#author's interleaving pading in github
    out2_4 = self.conv2_4(nn.functional.pad(x, (0, 1, 0, 1)))#right interleaving padding
    #out2\_4 = self.conv2\_4(nn.functional.pad(x, (1, 0, 1, 0)))#author's interleaving pading in github

    height = out1_1.size()[2]
    width = out1_1.size()[3]

    out1_1_2 = torch.stack((out1_1, out1_2), dim=-3).permute(0, 1, 3, 4, 2).contiguous().view(
        batch_size, -1, height, width \* 2)
    out1_3_4 = torch.stack((out1_3, out1_4), dim=-3).permute(0, 1, 3, 4, 2).contiguous().view(
        batch_size, -1, height, width \* 2)

    out1_1234 = torch.stack((out1_1_2, out1_3_4), dim=-3).permute(0, 1, 3, 2, 4).contiguous().view(
        batch_size, -1, height \* 2, width \* 2)

    out2_1_2 = torch.stack((out2_1, out2_2), dim=-3).permute(0, 1, 3, 4, 2).contiguous().view(
        batch_size, -1, height, width \* 2)
    out2_3_4 = torch.stack((out2_3, out2_4), dim=-3).permute(0, 1, 3, 4, 2).contiguous().view(
        batch_size, -1, height, width \* 2)

    out2_1234 = torch.stack((out2_1_2, out2_3_4), dim=-3).permute(0, 1, 3, 2, 4).contiguous().view(
        batch_size, -1, height \* 2, width \* 2)

    out1 = self.bn1_1(out1_1234)
    out1 = self.relu(out1)
    out1 = self.conv3(out1)
    out1 = self.bn2(out1)

    out2 = self.bn1_2(out2_1234)

    out = out1 + out2
    out = self.relu(out)

    return out

#编码，下采样
class Fcrn_encode(nn.Module):
def __init__(self, dim=opt.dim):
super(Fcrn_encode, self).init()
self.dim = dim
self.conv_1 = nn.Conv2d(in_channels=3, out_channels=dim, kernel_size=3, stride=1, padding=1)
self.residual_block_1_down_1 = ResidualBlockClass(‘Detector.Res1’, 1*dim, 2*dim, resample=‘down’, activate=‘leaky_relu’)

	# 128x128
    self.residual_block_2_down_1 = ResidualBlockClass('Detector.Res2', 2\*dim, 4\*dim, resample='down', activate='leaky\_relu')
	#64x64
    self.residual_block_3_down_1     = ResidualBlockClass('Detector.Res3', 4\*dim, 4\*dim, resample='down', activate='leaky\_relu')
	#32x32
    self.residual_block_4_down_1     = ResidualBlockClass('Detector.Res4', 4\*dim, 6\*dim, resample='down', activate='leaky\_relu')
	#16x16
    self.residual_block_5_none_1     = ResidualBlockClass('Detector.Res5', 6\*dim, 6\*dim, resample=None, activate='leaky\_relu')
    

def forward(self, x, n1=0, n2=0, n3=0):
    x1 = self.conv_1(x)#x1:dimx256x256
    x2 = self.residual_block_1_down_1(x1)#x2:2dimx128x128
    x3 = self.residual_block_2_down_1((1-opt.alpha)\*x2+opt.alpha\*n1)#x3:4dimx64x64
    x4 = self.residual_block_3_down_1((1-opt.alpha)\*x3+opt.alpha\*n2)#x4:4dimx32x32
    x = self.residual_block_4_down_1((1-opt.alpha)\*x4+opt.alpha\*n3)
    feature = self.residual_block_5_none_1(x)
    x = F.tanh(feature)       
    return x, x2, x3, x4

### 3.10、解码， 上采样



<