PointNet模型的Pytorch代码详解

前言

        关于PointNet模型的构成、原理、效果等等论文部分内容，我在之前一篇论文中写到过，可以参考这个链接：PointNet论文笔记    下边我就直接放一张网络组成图，并对代码进行解释，我以一种比较容易理解的顺序放置，希望耐心阅读。

 

网络结构图示

        在分类网络中，输入n个点，对输入做特征变换，再进行最大池化输出k个种类；分割网络是分类网络的一个拓展，它考虑了全局和局部的特征以及每个点的输出分数。mlp代表多层感知机，括号中是感知机的层数，批标准化（Batchnorm）本用于所有带有ReLU函数的层，Dropout层被用于分类网络中最后一个多层感知机中。
 

代码详解

        首先我先来讲解分类网络，图中深色部分，首先输入点经过一个transform，再经过多层感知机，再经过一个feature transform，再经过多层感知机和max pooling，最后经过多层感知机获得分类结果，网络结构是比较清晰的，下边一块一块看：

input transform

        首先这一层的目的是对输入的每一个点云，在这里是2500个三坐标点，目的是要获得一个3×3的变换矩阵，获得这个矩阵的原因是：要对点云的姿态进行校正，而该变换矩阵就是根据点云特性，做出一个刚体变换，使点云处于一个比较容易检测的姿态。先对输入经过三级卷积核为1×1的卷积处理得到1024通道的数据，再经过全连接处映射到九个数据，最后调整为3×3

class STN3d(nn.Module):
    def __init__(self):
        super(STN3d, self).__init__()
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, 9)
        self.relu = nn.ReLU()

        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)


    def forward(self, x):
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)

        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)

        iden = Variable(torch.from_numpy(np.array([1,0,0,0,1,0,0,0,1]).astype(np.float32))).view(1,9).repeat(batchsize,1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, 3, 3)
        return x

feature transform

        下边我们先考虑后边这个feature transform层，这个其实和上边那个是一样的，只是从电源数据中获取一个64×64的变换矩阵，这个也是对特征的一种校正，一种广义的位姿变换，代码几乎没有差别

class STNkd(nn.Module):
    def __init__(self, k=64):
        super(STNkd, self).__init__()
        self.conv1 = torch.nn.Conv1d(k, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k*k)
        self.relu = nn.ReLU()

        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.bn4 = nn.BatchNorm1d(512)
        self.bn5 = nn.BatchNorm1d(256)

        self.k = k

    def forward(self, x):
        batchsize = x.size()[0]
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)

        x = F.relu(self.bn4(self.fc1(x)))
        x = F.relu(self.bn5(self.fc2(x)))
        x = self.fc3(x)

        iden = Variable(torch.from_numpy(np.eye(self.k).flatten().astype(np.float32))).view(1,self.k*self.k).repeat(batchsize,1)
        if x.is_cuda:
            iden = iden.cuda()
        x = x + iden
        x = x.view(-1, self.k, self.k)
        return x

主体部分

        这部分讲max pooling之前的剩余部分，首先经过STN3d获得3×3矩阵，乘以点云做完位姿变换，再经过多层感知机（实际上多层感知机与卷积核边长为1的卷积操作本质是一样的），再乘以经过STNkd获得的64×64的矩阵，完成位姿变换，再经过多层感知机（这里同样用边长为1的卷积核的卷积操作），得到n×1024的矩阵，n为每批次读入的数据文件个数。下边这个类中调用了前边两个类。

class PointNetfeat(nn.Module):
    def __init__(self, global_feat = True, feature_transform = False):
        super(PointNetfeat, self).__init__()
        self.stn = STN3d()
        self.conv1 = torch.nn.Conv1d(3, 64, 1)
        self.conv2 = torch.nn.Conv1d(64, 128, 1)
        self.conv3 = torch.nn.Conv1d(128, 1024, 1)
        self.bn1 = nn.BatchNorm1d(64)
        self.bn2 = nn.BatchNorm1d(128)
        self.bn3 = nn.BatchNorm1d(1024)
        self.global_feat = global_feat
        self.feature_transform = feature_transform
        if self.feature_transform:
            self.fstn = STNkd(k=64)

    def forward(self, x):
        n_pts = x.size()[2]
        trans = self.stn(x)
        x = x.transpose(2, 1)
        x = torch.bmm(x, trans)
        x = x.transpose(2, 1)
        x = F.relu(self.bn1(self.conv1(x)))

        if self.feature_transform:
            trans_feat = self.fstn(x)
            x = x.transpose(2,1)
            x = torch.bmm(x, trans_feat)
            x = x.transpose(2,1)
        else:
            trans_feat = None

        pointfeat = x
        x = F.relu(self.bn2(self.conv2(x)))
        x = self.bn3(self.conv3(x))
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024)
        if self.global_feat:
            return x, trans, trans_feat
        else:
            x = x.view(-1, 1024, 1).repeat(1, 1, n_pts)
            return torch.cat([x, pointfeat], 1), trans, trans_feat

后处理部分

        下边就要进行最大池化和多层感知机进行分类了，经过全连接分成k类，根据概率来判别究竟属于哪一类

class PointNetCls(nn.Module):
    def __init__(self, k=2, feature_transform=False):
        super(PointNetCls, self).__init__()
        self.feature_transform = feature_transform
        self.feat = PointNetfeat(global_feat=True, feature_transform=feature_transform)
        self.fc1 = nn.Linear(1024, 512)
        self.fc2 = nn.Linear(512, 256)
        self.fc3 = nn.Linear(256, k)
        # 防止过拟合
        self.dropout = nn.Dropout(p=0.3)
        # 归一化防止梯度爆炸与梯度消失
        self.bn1 = nn.BatchNorm1d(512)
        self.bn2 = nn.BatchNorm1d(256)
        self.relu = nn.ReLU()

    def forward(self, x):
        # 完成网络主体部分
        x, trans, trans_feat = self.feat(x)
        # 经过三个全连接层（多层感知机）映射成k类
        x = F.relu(self.bn1(self.fc1(x)))
        x = F.relu(self.bn2(self.dropout(self.fc2(x))))
        x = self.fc3(x)
        # 返回的是该点云是第ki类的概率
        return F.log_softmax(x, dim=1), trans, trans_feat

 分割网络

        分割网络是借用了分类网络的两部分，分别是64通道和1024通道，堆积在一起形成1088通道的输入，经过多层感知机输出了结果m通道的结果，m代表类的个数，也就是每个点属于哪一类，实际上分割是在像素级或者点级的分类，本质上是一样的

class PointNetDenseCls(nn.Module):
    def __init__(self, k = 2, feature_transform=False):
        super(PointNetDenseCls, self).__init__()
        self.k = k
        self.feature_transform=feature_transform
        self.feat = PointNetfeat(global_feat=False, feature_transform=feature_transform)
        self.conv1 = torch.nn.Conv1d(1088, 512, 1)
        self.conv2 = torch.nn.Conv1d(512, 256, 1)
        self.conv3 = torch.nn.Conv1d(256, 128, 1)
        self.conv4 = torch.nn.Conv1d(128, self.k, 1)
        self.bn1 = nn.BatchNorm1d(512)
        self.bn2 = nn.BatchNorm1d(256)
        self.bn3 = nn.BatchNorm1d(128)

    def forward(self, x):
        batchsize = x.size()[0]
        n_pts = x.size()[2]
        x, trans, trans_feat = self.feat(x)
        x = F.relu(self.bn1(self.conv1(x)))
        x = F.relu(self.bn2(self.conv2(x)))
        x = F.relu(self.bn3(self.conv3(x)))
        x = self.conv4(x)
        x = x.transpose(2,1).contiguous()
        x = F.log_softmax(x.view(-1,self.k), dim=-1)
        x = x.view(batchsize, n_pts, self.k)
        return x, trans, trans_feat

训练结果

        训练过程可以参考源码

        网络分类性能还是很强的，只是迭代了一次，精度就达到了91%以上

                        

        在点较少的情况下，分割效果也还是可以的，5次迭代可以达到 80.0mIoU