点云论文

1. PointNet论文理解：

       https://blog.csdn.net/phosphenesvision/article/details/106724377?utm_medium=distribute.pc_relevant_t0.none-task-blog-BlogCommendFromMachineLearnPai2-1.channel_param&depth_1-utm_source=distribute.pc_relevant_t0.none-task-blog-BlogCommendFromMachineLearnPai2-1.channel_param

　　https://zhuanlan.zhihu.com/p/86331508

PointNet 源码理解：

      https://www.jianshu.com/p/4646016620db

      https://blog.csdn.net/cg129054036/article/details/105456002

      https://blog.csdn.net/jcsm__/article/details/109244748

import torch 
import numpy as np

class STN3d(nn.Module):
    '''
    3x3 transform
    '''
    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] #32
        x = F.relu(self.bn1(self.conv1(x))) #32 *64 *2500
        x = F.relu(self.bn2(self.conv2(x))) #32*128*2500
        x = F.relu(self.bn3(self.conv3(x))) #32*1024*2500
        x = torch.max(x, 2, keepdim=True)[0]
        x = x.view(-1, 1024) #32*1024
 
        x = F.relu(self.bn4(self.fc1(x))) #32*512
        x = F.relu(self.bn5(self.fc2(x))) #32*256
        x = self.fc3(x) #32*9
 
        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) #32*3*3
        return x
 
 
class STNkd(nn.Module):
    '''
    64x64 transform
    '''
    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
 
class PointNetfeat(nn.Module):
    '''
    Output: global feature / local+global feature
    '''
    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): #x ： 32*3*2500
        n_pts = x.size()[2] # 2500
        trans = self.stn(x) # 32*3*3
        x = x.transpose(2, 1) # 32*2500*3
        
        x = torch.bmm(x, trans) # X:32*3*2500  trans:32*2500*3  输出:32*3*3
        x = x.transpose(2, 1) # 32 *3*2500
        x = F.relu(self.bn1(self.conv1(x))) #32*64*2500
 
        if self.feature_transform:
            trans_feat = self.fstn(x) #输入x: 32*64*2500, 输出trans_feat：32*64*64
            x = x.transpose(2,1) #32*2500*64
            x = torch.bmm(x, trans_feat) #输入：32*2500*64 32*64*64 输出：32*2500*64
            x = x.transpose(2,1) #32*64*2500
        else:
            trans_feat = None
 
        pointfeat = x
        x = F.relu(self.bn2(self.conv2(x))) #32*128*2500
        x = self.bn3(self.conv3(x)) #32*1024*2500
        x = torch.max(x, 2, keepdim=True)[0] #32*1024的乘积
        x = x.view(-1, 1024) #32*1024

        if self.global_feat:
            return x, trans, trans_feat  # (B, 1024) (B, 3, 3) (B, 64, 64)
        else:
            x = x.view(-1, 1024, 1).repeat(1, 1, n_pts)
            return torch.cat([x, pointfeat], 1), trans, trans_feat # (B, 1088, 2500) (B,3, 3) (B, 64, 64)
 
class PointNetCls(nn.Module):
    # 分类网络
    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)
        self.dropout = nn.Dropout(p=0.3)
        self.bn1 = nn.BatchNorm1d(512)
        self.bn2 = nn.BatchNorm1d(256)
        self.relu = nn.ReLU()
 
    def forward(self, x):
        x, trans, trans_feat = self.feat(x) #32*1024 32*3*3 32*64*64
        x = F.relu(self.bn1(self.fc1(x))) #32*512
        x = F.relu(self.bn2(self.dropout(self.fc2(x)))) #32*256
        x = self.fc3(x) #32*K
        return F.log_softmax(x, dim=1), trans, trans_feat
 
 
class PointNetDenseCls(nn.Module):
    # 分割网络
    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):  #32*3*2500
        batchsize = x.size()[0]  #32
        n_pts = x.size()[2] #2500
        x, trans, trans_feat = self.feat(x) #[32, 1088, 2500]

        x = F.relu(self.bn1(self.conv1(x))) #[32, 512, 2500]
        x = F.relu(self.bn2(self.conv2(x))) #[32, 256, 2500]
        x = F.relu(self.bn3(self.conv3(x))) #[32, 128, 2500]
        x = self.conv4(x)#[32, K, 2500]
        x = x.transpose(2,1).contiguous()#[32*2500*k]
        x = F.log_softmax(x.view(-1,self.k), dim=-1)
        x = x.view(batchsize, n_pts, self.k)#[32, 2500, k]
        return x, trans, trans_feat
 
if __name__ == '__main__':
    sim_data = Variable(torch.rand(32,3,2500))
    trans = STN3d()
    out = trans(sim_data)
    print('stn', out.size())
    print('loss', feature_transform_regularizer(out))
 
    sim_data_64d = Variable(torch.rand(32, 64, 2500))
    trans = STNkd(k=64)
    out = trans(sim_data_64d)
    print('stn64d', out.size())
    print('loss', feature_transform_regularizer(out))
 
    pointfeat = PointNetfeat(global_feat=True)
    out, _, _ = pointfeat(sim_data)
    print('global feat', out.size())
 
    pointfeat = PointNetfeat(global_feat=False)
    out, _, _ = pointfeat(sim_data)
    print('point feat', out.size())
 
    cls = PointNetCls(k = 5)
    out, _, _ = cls(sim_data)
    print('class', out.size())
 
    seg = PointNetDenseCls(k = 3)
    out, _, _ = seg(sim_data)
    print('seg', out.size())

2. PointNet++论文理解：

      https://blog.csdn.net/yong_qi2015/article/details/108957905

　  https://zhuanlan.zhihu.com/p/88238420

源码理解：

      https://blog.csdn.net/cg129054036/article/details/105545895?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-2.channel_param&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-2.channel_param