卷积定义

class Conv3(nn.Module):def __init__(self, in_ch, out_ch, kernel_size=3,stride=2,padding=1):super(Conv3, self).__init__()self.conv = nn.Sequential(nn.Conv3d(in_ch, out_ch, kernel_size=kernel_size, stride=stride,padding=padding),nn.BatchNorm3d(out_ch),  # 添加了BN层nn.LeakyReLU())def forward(self, input):return self.conv(input)class Dconv2(nn.Module):def __init__(self, in_ch, out_ch, kernel_size=3,padding=1,output_padding=1,stride =2):super(Dconv2, self).__init__()self.convtrans = nn.Sequential(nn.ConvTranspose2d(in_ch, out_ch, kernel_size=kernel_size,stride=stride,padding=padding,output_padding=output_padding),nn.BatchNorm2d(out_ch),  # 添加了BN层nn.ReLU())def forward(self, input):return self.convtrans(input)class Conv2(nn.Module):def __init__(self, in_ch, out_ch, kernel_size=2,stride=2,padding=0):super(Conv2, self).__init__()self.conv = nn.Sequential(nn.Conv2d(in_ch, out_ch, kernel_size=kernel_size, stride=stride,padding=padding),nn.BatchNorm2d(out_ch),  # 添加了BN层nn.LeakyReLU())def forward(self, input):return self.conv(input)

生成器


class Generator(nn.Module):def __init__(self):super(Generator, self).__init__()self.encoder = nn.Sequential( #input  [1, 64, 32, 96]Conv3(1,16),        # [16, 32, 16, 48]Conv3(16, 64),      # [64, 16, 8, 24]Conv3(64, 256),     # [256, 8, 4, 12]Conv3(256, 512),    # [512, 4, 2, 6]Conv3(512, 1024),   # [1024, 2, 1, 3]# Conv3(1024, 2048,kernel_size=(2,1,3),stride=1),  # [2048, 1, 1, 1]nn.Conv3d(1024, 2048, kernel_size=(2, 1, 3), stride=(1, 1, 1)),nn.Tanh())self.decoder = nn.Sequential(Dconv2(2048, 1024, kernel_size=(2, 1), stride=2,output_padding=(0,0),padding=0),Dconv2(1024, 512,  stride=2),Dconv2(512, 256, stride=2, ),Dconv2(256, 128,  stride=2),Dconv2(128, 64,  stride=2),Dconv2(64, 32,  stride=2),)self.decoder_ = nn.Sequential(Dconv2(40, 16 ),Dconv2( 16,1),nn.Conv2d(1,1,1,1))def forward(self, input):x = self.encoder(input)x = torch.squeeze(x, 4)x = self.decoder(x)skip, _ = torch.topk(input, 8, dim=4, largest=True)skip = skip.transpose(4,1)skip = skip.squeeze(-1)x = torch.cat([skip, x], dim=1)x = self.decoder_(x)# print("x",x.size())x = x.transpose(3,2)return x

鉴别器

class Discriminator(nn.Module):def __init__(self):super(Discriminator, self).__init__()self.ThreeEncoder = nn.Sequential( #input  [1, 64, 32, 96]Conv3(1,16),        # [16, 32, 16, 48]Conv3(16, 64),      # [64, 16, 8, 24]Conv3(64, 256),     # [256, 8, 4, 12]Conv3(256, 512),    # [512, 4, 2, 6]Conv3(512, 1024),   # [1024, 2, 1, 3]nn.Conv3d(1024, 2048, kernel_size=(2, 1, 3), stride=(1, 1, 1)),nn.ReLU())self.conv1 = nn.Conv2d(2048, 512, kernel_size=1)self.relu = nn.ReLU()self.TowEncoder  = nn.Sequential( #input  [1, 256,128]Conv2(1, 4, kernel_size=2, stride=2),Conv2(4, 8, kernel_size=2, stride=2),Conv2(8, 16, kernel_size=2, stride=2),Conv2(16, 32, kernel_size=2, stride=2),Conv2(32, 64, kernel_size=2, stride=2),Conv2(64, 128, kernel_size=2, stride=2),Conv2(128, 256, kernel_size=2, stride=2),# Conv2(256, 512, kernel_size=1, stride=2),nn.Conv2d(256, 512, kernel_size=1, stride=2),nn.ReLU(),)self.fc = nn.Sequential(nn.Linear(1024, 256),nn.ReLU(),nn.Linear(256, 1),nn.Sigmoid())def forward(self, input3d, input2d):x = self.ThreeEncoder(input3d)x = torch.squeeze(x, -1)x = self.conv1(x)x = self.relu(x)y = self.TowEncoder(input2d)x = torch.cat([x, y], dim=1)x = x.view(x.size()[0], -1)x = self.fc(x)return x

测试

if __name__ == '__main__':i = torch.randn( 1,1, 64, 32, 96)g = Generator()output = g(i)print(output.shape)j = torch.randn(1, 1, 128, 256)d = Discriminator()out = d(i,j)print(out.shape)

pipline

见github吧，写的一般，就不搬上来了

效果

可能是参数调的不好，意思到了。

论文 Through Fog High Resolution Imaging Using Millimeter Wave Radar 复现相关推荐

论文笔记《Spatially Sparse Precoding in Millimeter Wave MIMO Systems》
论文笔记<Spatially Sparse Precoding in Millimeter Wave MIMO Systems> 背景增加商业无线系统可用的频谱,通过探索新的不那么受限的 ...
Retrieval of ice cloud microphysical parameters using the CloudSat millimeter‐wave radar and tempera
@[TOC]Retrieval of ice cloud microphysical parameters using the CloudSat millimeter‐wave radar and t ...
深度学习论文: Computer Vision for Road Imaging and Pothole Detection: A State-of-the-Art Review
深度学习论文: Computer Vision for Road Imaging and Pothole Detection: A State-of-the-Art Review of Systems ...
【论文阅读】An Image Patch is a Wave: Quantum Inspired Vision MLP
[论文阅读]An Image Patch is a Wave: Quantum Inspired Vision MLP 1.摘要 2.视觉领域的几种研究架构 2.1CNN 2.2Transformer ...
论文：EMNLP2021-End-to-End Entity Resolution and Question Answering Using Differentiable Knowledge Grap
何为实体解析(entity resolution):entity resolution component identifies entities mentioned in the question, ...
论文阅读：RA-Depth: Resolution Adaptive Self-Supervised Monocular Depth Estimation
中文标题: 分辨自适应自监督单目深度估计提出问题传统的无监督深度估计方法如果在固定分辨率上训练,则迁移到其他分辨率上存在严重退化. 创新点提出一种分辨率自适应的无监督深度学习框架(RA-Dept ...
[论文笔记] Super Low Resolution RF Powered Accelerometers for Alerting on Hospitalized Patient Bed Exits
原文链接:Super Low Resolution RF Powered Accelerometers for Alerting on Hospitalized Patient Bed Exits | ...
毫米波点云生成论文阅读笔记 | 3D Point Cloud Generation with Millimeter-Wave Radar
毫米波点云生成论文 | 3D Point Cloud Generation with Millimeter-Wave Radar Kun Qian, Zhaoyuan He, Xinyu Zhang ...
【AAAI 2020】Gait Recognition for Co-Existing Multiple People Using Millimeter Wave Sensing
文章目录 1. 四个问题 1. 解决什么问题 2. 用了什么方法解决 3. 效果如何 4. 还存在什么问题 2. 论文介绍 3. 参考资料 4. 收获 1. 四个问题 1. 解决什么问题步态识别,根 ...

论文 Through Fog High Resolution Imaging Using Millimeter Wave Radar 复现

Through Fog High Resolution Imaging Using Millimeter Wave Radar

卷积定义

生成器

鉴别器

测试

pipline

效果

论文 Through Fog High Resolution Imaging Using Millimeter Wave Radar 复现相关推荐

最新文章

热门文章