PointNet

最新推荐文章于 2024-07-09 16:26:08 发布
最新推荐文章于 2024-07-09 16:26:08 发布
阅读量129 收藏
点赞数
分类专栏： pointnet
原文链接：https://blog.csdn.net/moumde/article/details/106497764?spm=1001.2014.3001.5501
版权

PointNet代码详解

最近在做点云深度学习的机器人抓取，这篇博客主要是把近期学习PointNet的一些总结的知识点汇总一下。

PointNet概述详见以下网址和博客，这里也就不再赘述了。
三维深度学习之pointnet系列详解
PointNet网络结构详细解析
PointNet论文理解和代码分析
PointNet论文复现及代码详解

这里着重来探讨一下内部的代码（pointnet-master\models路径下的）。
PointNet原文及Github代码下载

详细的网络结构图如下
在这里插入图片描述
主要讲一下应该注意的地方：
（1）网络结构内部主要分为分类和分割两部分，从 global feature 开始区分分类和分割，关于点云的分类和分割，详见点云分类与分割的区别联系。
（2）我们这边主要定义数据维度的表示为 (B, H, W, C) ，也就是Batch, Height, Width, Channel。 开始输入时是一个3D的张量 (B, n, 3)，其中B即为训练的批量， n 为点云个数，3则代表了点云的(x,y,z)的3个位置，因此为了后续的卷积操作，会将其增加维度到4D张量(B, n, 3, 1)，方便后面卷积核提取产生特征通道数C，(B, n, 3, C)。
（3）第一层的卷积核大小为(1, 3)，因为每个点的维度都是(x, y, z)，后续的所有卷积核大小均为(1, 1)，因为经过第一次卷积之后数据就变为了(B, n, 1, C)。
（4）整个网络框架内部使用了两个分支网络Transform(T-Net)。T-Net对原样本进行一定的卷积和全连接等操作后得到变换矩阵并与原样本相乘完成点云内部结构的调整，但并不改变原样本数据的格式。第一次T-Net输出一个33的矩阵，第二次T-Net输出一个6464的矩阵。

阅读后续代码前请仔细看完这篇博客，务必理解里面内容
PointNet网络结构详细解析

T-Net

对应文件为“pointnet-master\models\transform_nets.py”
根据网络结构图可知输入量时B×n×3，对于input_transform来说，主要经历了以下处理过程：
卷积：64–128–1024
全连接：1024–512–256–3*K（代码中给出K=3）
最后reshape得到变换矩阵

def input_transform_net(point_cloud, is_training, bn_decay=None, K=3):
    """ Input (XYZ) Transform Net, input is BxNx3 gray image
        Return:
            Transformation matrix of size 3xK """
    #建议阅读时忽略batch_size的维度，将张量视作一个3维矩阵[n, W, C]
    #其中n为点云数，也就是Height；W为矩阵宽Width；C为特征通道数
    batch_size = point_cloud.get_shape()[0].value	#得到训练批量
    num_point = point_cloud.get_shape()[1].value	#得到点云个数

input_image <span class="token operator">=</span> tf<span class="token punctuation">.</span>expand_dims<span class="token punctuation">(</span>point_cloud<span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">)</span>	<span class="token comment">#扩展维度为四维张量，-1表示最后一个维度，第四个维度来表示特征通道数</span>
<span class="token comment">#input_image  [batch_size, num_point, 3, 1]</span>
<span class="token comment">#第一次卷积，采用64个大小为[1,3]的卷积核</span>
<span class="token comment">#完成之后  net表示为[batch_size, num_point, 1, 64]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>input_image<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">3</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'tconv1'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
<span class="token comment">#第二次卷积，采用128个大小为[1,1]的卷积核</span>
<span class="token comment">#完成之后  net表示为[batch_size, num_point, 1, 128]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">128</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'tconv2'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
<span class="token comment">#第三次卷积，采用1024个大小为[1,1]的卷积核</span>
<span class="token comment">#完成之后  net表示为[batch_size, num_point, 1, 1024]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">1024</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'tconv3'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
<span class="token comment">#max pooling，扫描的模板大小为[num_point, 1]，也就是每一个特征通道仅保留一个feature</span>
<span class="token comment">#完成该池化操作之后的net表示为[batch_size, 1, 1, 1024]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>max_pool2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span>num_point<span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                         padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> scope<span class="token operator">=</span><span class="token string">'tmaxpool'</span><span class="token punctuation">)</span>
<span class="token comment">#reshape张量，即将其平面化为[batch_size, 1024]</span>
net <span class="token operator">=</span> tf<span class="token punctuation">.</span>reshape<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span>batch_size<span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>fully_connected<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">512</span><span class="token punctuation">,</span> bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                              scope<span class="token operator">=</span><span class="token string">'tfc1'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>fully_connected<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">256</span><span class="token punctuation">,</span> bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                              scope<span class="token operator">=</span><span class="token string">'tfc2'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>

<span class="token comment">#最后得到的net表示为[batch_size, 256]，即每组只保留256个特征</span>
<span class="token comment">#再经过下述的全连接操作之后得到[batch_size, 3*K]大小的transform</span>
<span class="token keyword">with</span> tf<span class="token punctuation">.</span>variable_scope<span class="token punctuation">(</span><span class="token string">'transform_XYZ'</span><span class="token punctuation">)</span> <span class="token keyword">as</span> sc<span class="token punctuation">:</span>
    <span class="token keyword">assert</span><span class="token punctuation">(</span>K<span class="token operator">==</span><span class="token number">3</span><span class="token punctuation">)</span>
    weights <span class="token operator">=</span> tf<span class="token punctuation">.</span>get_variable<span class="token punctuation">(</span><span class="token string">'weights'</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">256</span><span class="token punctuation">,</span> <span class="token number">3</span><span class="token operator">*</span>K<span class="token punctuation">]</span><span class="token punctuation">,</span>
                              initializer<span class="token operator">=</span>tf<span class="token punctuation">.</span>constant_initializer<span class="token punctuation">(</span><span class="token number">0.0</span><span class="token punctuation">)</span><span class="token punctuation">,</span>
                              dtype<span class="token operator">=</span>tf<span class="token punctuation">.</span>float32<span class="token punctuation">)</span>
    biases <span class="token operator">=</span> tf<span class="token punctuation">.</span>get_variable<span class="token punctuation">(</span><span class="token string">'biases'</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">3</span><span class="token operator">*</span>K<span class="token punctuation">]</span><span class="token punctuation">,</span>
                             initializer<span class="token operator">=</span>tf<span class="token punctuation">.</span>constant_initializer<span class="token punctuation">(</span><span class="token number">0.0</span><span class="token punctuation">)</span><span class="token punctuation">,</span>
                             dtype<span class="token operator">=</span>tf<span class="token punctuation">.</span>float32<span class="token punctuation">)</span>
    biases <span class="token operator">+=</span> tf<span class="token punctuation">.</span>constant<span class="token punctuation">(</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">0</span><span class="token punctuation">,</span><span class="token number">0</span><span class="token punctuation">,</span><span class="token number">0</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">0</span><span class="token punctuation">,</span><span class="token number">0</span><span class="token punctuation">,</span><span class="token number">0</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span> dtype<span class="token operator">=</span>tf<span class="token punctuation">.</span>float32<span class="token punctuation">)</span>
    transform <span class="token operator">=</span> tf<span class="token punctuation">.</span>matmul<span class="token punctuation">(</span>net<span class="token punctuation">,</span> weights<span class="token punctuation">)</span>
    transform <span class="token operator">=</span> tf<span class="token punctuation">.</span>nn<span class="token punctuation">.</span>bias_add<span class="token punctuation">(</span>transform<span class="token punctuation">,</span> biases<span class="token punctuation">)</span>

<span class="token comment">#重新塑造变换矩阵[batch_size, 3*K]为[batch_size, 3, K]</span>
transform <span class="token operator">=</span> tf<span class="token punctuation">.</span>reshape<span class="token punctuation">(</span>transform<span class="token punctuation">,</span> <span class="token punctuation">[</span>batch_size<span class="token punctuation">,</span> <span class="token number">3</span><span class="token punctuation">,</span> K<span class="token punctuation">]</span><span class="token punctuation">)</span>
<span class="token keyword">return</span> transform
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
  • 47
  • 48
  • 49
  • 50
  • 51
  • 52
  • 53
  • 54
  • 55
  • 56
  • 57

feature_transform同input_transform类似，经历处理过程表示为：
卷积：64–128–1024
全连接：1024–512–256–64*K（代码中给出K=64）
最后reshape得到变换矩阵
这里就不再注释，大家也可以通过下面未注释的代码测试一下上一个transform有没有看懂了。

def feature_transform_net(inputs, is_training, bn_decay=None, K=64):
    """ Feature Transform Net, input is BxNx1xK
        Return:
            Transformation matrix of size KxK """
    batch_size = inputs.get_shape()[0].value
    num_point = inputs.get_shape()[1].value

net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>inputs<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'tconv1'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">128</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'tconv2'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">1024</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'tconv3'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>max_pool2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span>num_point<span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                         padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> scope<span class="token operator">=</span><span class="token string">'tmaxpool'</span><span class="token punctuation">)</span>

net <span class="token operator">=</span> tf<span class="token punctuation">.</span>reshape<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span>batch_size<span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>fully_connected<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">512</span><span class="token punctuation">,</span> bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                              scope<span class="token operator">=</span><span class="token string">'tfc1'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>fully_connected<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">256</span><span class="token punctuation">,</span> bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                              scope<span class="token operator">=</span><span class="token string">'tfc2'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>

<span class="token keyword">with</span> tf<span class="token punctuation">.</span>variable_scope<span class="token punctuation">(</span><span class="token string">'transform_feat'</span><span class="token punctuation">)</span> <span class="token keyword">as</span> sc<span class="token punctuation">:</span>
    weights <span class="token operator">=</span> tf<span class="token punctuation">.</span>get_variable<span class="token punctuation">(</span><span class="token string">'weights'</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">256</span><span class="token punctuation">,</span> K<span class="token operator">*</span>K<span class="token punctuation">]</span><span class="token punctuation">,</span>
                              initializer<span class="token operator">=</span>tf<span class="token punctuation">.</span>constant_initializer<span class="token punctuation">(</span><span class="token number">0.0</span><span class="token punctuation">)</span><span class="token punctuation">,</span>
                              dtype<span class="token operator">=</span>tf<span class="token punctuation">.</span>float32<span class="token punctuation">)</span>
    biases <span class="token operator">=</span> tf<span class="token punctuation">.</span>get_variable<span class="token punctuation">(</span><span class="token string">'biases'</span><span class="token punctuation">,</span> <span class="token punctuation">[</span>K<span class="token operator">*</span>K<span class="token punctuation">]</span><span class="token punctuation">,</span>
                             initializer<span class="token operator">=</span>tf<span class="token punctuation">.</span>constant_initializer<span class="token punctuation">(</span><span class="token number">0.0</span><span class="token punctuation">)</span><span class="token punctuation">,</span>
                             dtype<span class="token operator">=</span>tf<span class="token punctuation">.</span>float32<span class="token punctuation">)</span>
    biases <span class="token operator">+=</span> tf<span class="token punctuation">.</span>constant<span class="token punctuation">(</span>np<span class="token punctuation">.</span>eye<span class="token punctuation">(</span>K<span class="token punctuation">)</span><span class="token punctuation">.</span>flatten<span class="token punctuation">(</span><span class="token punctuation">)</span><span class="token punctuation">,</span> dtype<span class="token operator">=</span>tf<span class="token punctuation">.</span>float32<span class="token punctuation">)</span>
    transform <span class="token operator">=</span> tf<span class="token punctuation">.</span>matmul<span class="token punctuation">(</span>net<span class="token punctuation">,</span> weights<span class="token punctuation">)</span>
    transform <span class="token operator">=</span> tf<span class="token punctuation">.</span>nn<span class="token punctuation">.</span>bias_add<span class="token punctuation">(</span>transform<span class="token punctuation">,</span> biases<span class="token punctuation">)</span>

transform <span class="token operator">=</span> tf<span class="token punctuation">.</span>reshape<span class="token punctuation">(</span>transform<span class="token punctuation">,</span> <span class="token punctuation">[</span>batch_size<span class="token punctuation">,</span> K<span class="token punctuation">,</span> K<span class="token punctuation">]</span><span class="token punctuation">)</span>
<span class="token keyword">return</span> transform
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41

分类网络结构

分类网络内容即PointNet网络结构图中最上面的那个主要框图，即Classification Network。对应文件为“pointnet-master\models\pointnet_cls.py”

def get_model(point_cloud, is_training, bn_decay=None):
    """ Classification PointNet, input is BxNx3, output Bx40 """
    batch_size = point_cloud.get_shape()[0].value	#得到训练批量
    num_point = point_cloud.get_shape()[1].value	#得到点云个数
    end_points = {}									#生成字典

<span class="token keyword">with</span> tf<span class="token punctuation">.</span>variable_scope<span class="token punctuation">(</span><span class="token string">'transform_net1'</span><span class="token punctuation">)</span> <span class="token keyword">as</span> sc<span class="token punctuation">:</span>
    transform <span class="token operator">=</span> input_transform_net<span class="token punctuation">(</span>point_cloud<span class="token punctuation">,</span> is_training<span class="token punctuation">,</span> bn_decay<span class="token punctuation">,</span> K<span class="token operator">=</span><span class="token number">3</span><span class="token punctuation">)</span>	<span class="token comment">#通过第一个T-Net得到input_tranform</span>
point_cloud_transformed <span class="token operator">=</span> tf<span class="token punctuation">.</span>matmul<span class="token punctuation">(</span>point_cloud<span class="token punctuation">,</span> transform<span class="token punctuation">)</span>	<span class="token comment">#将原矩阵和input_transform相乘完成转换</span>
input_image <span class="token operator">=</span> tf<span class="token punctuation">.</span>expand_dims<span class="token punctuation">(</span>point_cloud_transformed<span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">)</span>	<span class="token comment">#扩展为4维张量[batch_size, num_point, 3, 1]</span>

<span class="token comment">#使用64个大小为[1,3]的卷积核得到64个特征通道</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>input_image<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">3</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv1'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
<span class="token comment">#使用64个大小为[1,1]的卷积核得到64个特征通道</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv2'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
<span class="token comment">#最后得到net为[batch_size, point_num, 1, 64]</span>
<span class="token comment">#第二次使用T-Net进行转换</span>
<span class="token keyword">with</span> tf<span class="token punctuation">.</span>variable_scope<span class="token punctuation">(</span><span class="token string">'transform_net2'</span><span class="token punctuation">)</span> <span class="token keyword">as</span> sc<span class="token punctuation">:</span>
    transform <span class="token operator">=</span> feature_transform_net<span class="token punctuation">(</span>net<span class="token punctuation">,</span> is_training<span class="token punctuation">,</span> bn_decay<span class="token punctuation">,</span> K<span class="token operator">=</span><span class="token number">64</span><span class="token punctuation">)</span>
end_points<span class="token punctuation">[</span><span class="token string">'transform'</span><span class="token punctuation">]</span> <span class="token operator">=</span> transform		<span class="token comment">#在字典中用transform保存feature transform，记录原始特征</span>
<span class="token comment">#squeeze操作删除大小为1的维度</span>
<span class="token comment">#注意到之前操作之后得到的net为[batch_size, num_point, 1, 64]</span>
<span class="token comment">#从维度0开始，这里指定为第二个维度，squeeze之后为[batch_size, num_point, 64]</span>
<span class="token comment">#然后相乘得到变换后的矩阵</span>
net_transformed <span class="token operator">=</span> tf<span class="token punctuation">.</span>matmul<span class="token punctuation">(</span>tf<span class="token punctuation">.</span>squeeze<span class="token punctuation">(</span>net<span class="token punctuation">,</span> axis<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">2</span><span class="token punctuation">]</span><span class="token punctuation">)</span><span class="token punctuation">,</span> transform<span class="token punctuation">)</span>
<span class="token comment">#指定第二个维度并膨胀为4维张量[batch_size, num_point, 1, 64]</span>
net_transformed <span class="token operator">=</span> tf<span class="token punctuation">.</span>expand_dims<span class="token punctuation">(</span>net_transformed<span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">2</span><span class="token punctuation">]</span><span class="token punctuation">)</span>

<span class="token comment">#进行3次卷积操作，最后得到[batch_size, num_point, 1, 1024]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net_transformed<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv3'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">128</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv4'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">1024</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv5'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>

<span class="token comment"># Symmetric function: max pooling</span>
<span class="token comment">#max pooling，扫描的模板大小为[num_point, 1]，也就是每一个特征通道仅保留一个feature</span>
<span class="token comment">#完成该池化操作之后的net表示为[batch_size, 1, 1, 1024]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>max_pool2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span>num_point<span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                         padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> scope<span class="token operator">=</span><span class="token string">'maxpool'</span><span class="token punctuation">)</span>

<span class="token comment">#重塑张量得到[batch_size, 1024]</span>
net <span class="token operator">=</span> tf<span class="token punctuation">.</span>reshape<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span>batch_size<span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span>
<span class="token comment">#3次全连接并且进行2次dropout</span>
<span class="token comment">#最后得到40种分类结果</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>fully_connected<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">512</span><span class="token punctuation">,</span> bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                              scope<span class="token operator">=</span><span class="token string">'fc1'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>dropout<span class="token punctuation">(</span>net<span class="token punctuation">,</span> keep_prob<span class="token operator">=</span><span class="token number">0.7</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                      scope<span class="token operator">=</span><span class="token string">'dp1'</span><span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>fully_connected<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">256</span><span class="token punctuation">,</span> bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                              scope<span class="token operator">=</span><span class="token string">'fc2'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>dropout<span class="token punctuation">(</span>net<span class="token punctuation">,</span> keep_prob<span class="token operator">=</span><span class="token number">0.7</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                      scope<span class="token operator">=</span><span class="token string">'dp2'</span><span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>fully_connected<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">40</span><span class="token punctuation">,</span> activation_fn<span class="token operator">=</span><span class="token boolean">None</span><span class="token punctuation">,</span> scope<span class="token operator">=</span><span class="token string">'fc3'</span><span class="token punctuation">)</span>

<span class="token comment">#return返回分类结果以及n*64的原始特征</span>
<span class="token keyword">return</span> net<span class="token punctuation">,</span> end_points
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
  • 47
  • 48
  • 49
  • 50
  • 51
  • 52
  • 53
  • 54
  • 55
  • 56
  • 57
  • 58
  • 59
  • 60
  • 61
  • 62
  • 63
  • 64
  • 65
  • 66
  • 67
  • 68
  • 69
  • 70

分割网络结构

在PointNet原网络结构图中从global_feature中向下的分支，即Segmentation Network。对应文件为“pointnet-master\models\pointnet_seg.py”。

由于结构与分类网络类似，这里仅简单注释一下。

def get_model(point_cloud, is_training, bn_decay=None):
    """ Classification PointNet, input is BxNx3, output BxNx50 """
    batch_size = point_cloud.get_shape()[0].value
    num_point = point_cloud.get_shape()[1].value
    end_points = {}

<span class="token comment">#第一次T-Net转换，转换完成之后膨胀为四维张量</span>
<span class="token keyword">with</span> tf<span class="token punctuation">.</span>variable_scope<span class="token punctuation">(</span><span class="token string">'transform_net1'</span><span class="token punctuation">)</span> <span class="token keyword">as</span> sc<span class="token punctuation">:</span>
    transform <span class="token operator">=</span> input_transform_net<span class="token punctuation">(</span>point_cloud<span class="token punctuation">,</span> is_training<span class="token punctuation">,</span> bn_decay<span class="token punctuation">,</span> K<span class="token operator">=</span><span class="token number">3</span><span class="token punctuation">)</span>
point_cloud_transformed <span class="token operator">=</span> tf<span class="token punctuation">.</span>matmul<span class="token punctuation">(</span>point_cloud<span class="token punctuation">,</span> transform<span class="token punctuation">)</span>
input_image <span class="token operator">=</span> tf<span class="token punctuation">.</span>expand_dims<span class="token punctuation">(</span>point_cloud_transformed<span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">)</span>

<span class="token comment">#两次卷积，完成后得到[batch_size, num_point, 1, 64]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>input_image<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">3</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv1'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv2'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>

<span class="token comment">#第二次T-Net转换，转换过程中注意维度的变换操作</span>
<span class="token keyword">with</span> tf<span class="token punctuation">.</span>variable_scope<span class="token punctuation">(</span><span class="token string">'transform_net2'</span><span class="token punctuation">)</span> <span class="token keyword">as</span> sc<span class="token punctuation">:</span>
    transform <span class="token operator">=</span> feature_transform_net<span class="token punctuation">(</span>net<span class="token punctuation">,</span> is_training<span class="token punctuation">,</span> bn_decay<span class="token punctuation">,</span> K<span class="token operator">=</span><span class="token number">64</span><span class="token punctuation">)</span>
end_points<span class="token punctuation">[</span><span class="token string">'transform'</span><span class="token punctuation">]</span> <span class="token operator">=</span> transform
net_transformed <span class="token operator">=</span> tf<span class="token punctuation">.</span>matmul<span class="token punctuation">(</span>tf<span class="token punctuation">.</span>squeeze<span class="token punctuation">(</span>net<span class="token punctuation">,</span> axis<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">2</span><span class="token punctuation">]</span><span class="token punctuation">)</span><span class="token punctuation">,</span> transform<span class="token punctuation">)</span>
point_feat <span class="token operator">=</span> tf<span class="token punctuation">.</span>expand_dims<span class="token punctuation">(</span>net_transformed<span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">2</span><span class="token punctuation">]</span><span class="token punctuation">)</span>
<span class="token keyword">print</span><span class="token punctuation">(</span>point_feat<span class="token punctuation">)</span>

<span class="token comment">#3次卷积操作和1次最大池化操作</span>
<span class="token comment">#最后得到[batch_size, 1, 1, 1024]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>point_feat<span class="token punctuation">,</span> <span class="token number">64</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv3'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">128</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv4'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">1024</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv5'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
global_feat <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>max_pool2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span>num_point<span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                                 padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> scope<span class="token operator">=</span><span class="token string">'maxpool'</span><span class="token punctuation">)</span>
<span class="token keyword">print</span><span class="token punctuation">(</span>global_feat<span class="token punctuation">)</span>

<span class="token comment">#未进行性tile之前为[batch_size, 1, 1, 1024]</span>
<span class="token comment">#tile()函数用来对张量进行扩展，对维度1复制拓展至num_point倍</span>
<span class="token comment">#tile()操作之后为[batch_size, num_point, 1, 1024]</span>
global_feat_expand <span class="token operator">=</span> tf<span class="token punctuation">.</span>tile<span class="token punctuation">(</span>global_feat<span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span> num_point<span class="token punctuation">,</span> <span class="token number">1</span><span class="token punctuation">,</span> <span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span>
<span class="token comment">#tf.concat()将两个矩阵进行拼接，拼接得到网络结构中的n×1088的矩阵，n即num_point，此处忽略batch_size</span>
concat_feat <span class="token operator">=</span> tf<span class="token punctuation">.</span>concat<span class="token punctuation">(</span><span class="token number">3</span><span class="token punctuation">,</span> <span class="token punctuation">[</span>point_feat<span class="token punctuation">,</span> global_feat_expand<span class="token punctuation">]</span><span class="token punctuation">)</span>
<span class="token keyword">print</span><span class="token punctuation">(</span>concat_feat<span class="token punctuation">)</span>

<span class="token comment">#5次卷积操作得到128个特征通道输出 [batch_size, num_point, 1, 128]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>concat_feat<span class="token punctuation">,</span> <span class="token number">512</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv6'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">256</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv7'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">128</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv8'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">128</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     bn<span class="token operator">=</span><span class="token boolean">True</span><span class="token punctuation">,</span> is_training<span class="token operator">=</span>is_training<span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv9'</span><span class="token punctuation">,</span> bn_decay<span class="token operator">=</span>bn_decay<span class="token punctuation">)</span>
<span class="token comment">#max_pooling操作得到50个特征输出</span>
<span class="token comment">#[batch_size, num_point, 1, 50]</span>
net <span class="token operator">=</span> tf_util<span class="token punctuation">.</span>conv2d<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token number">50</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span>
                     padding<span class="token operator">=</span><span class="token string">'VALID'</span><span class="token punctuation">,</span> stride<span class="token operator">=</span><span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">,</span> activation_fn<span class="token operator">=</span><span class="token boolean">None</span><span class="token punctuation">,</span>
                     scope<span class="token operator">=</span><span class="token string">'conv10'</span><span class="token punctuation">)</span>
<span class="token comment">#squeeze()删除大小为1的第二个维度</span>
net <span class="token operator">=</span> tf<span class="token punctuation">.</span>squeeze<span class="token punctuation">(</span>net<span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">2</span><span class="token punctuation">]</span><span class="token punctuation">)</span> <span class="token comment"># BxNxC</span>

<span class="token keyword">return</span> net<span class="token punctuation">,</span> end_points
  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27
  • 28
  • 29
  • 30
  • 31
  • 32
  • 33
  • 34
  • 35
  • 36
  • 37
  • 38
  • 39
  • 40
  • 41
  • 42
  • 43
  • 44
  • 45
  • 46
  • 47
  • 48
  • 49
  • 50
  • 51
  • 52
  • 53
  • 54
  • 55
  • 56
  • 57
  • 58
  • 59
  • 60
  • 61
  • 62
  • 63
  • 64
  • 65
  • 66
  • 67
  • 68
  • 69
  • 70
  • 71
  • 72
  • 73
  • 74
  • 75
  • 76
  • 77
  • 78
  • 79
  • 80
  • 81
  • 82

以上便是此次全部代码解释内容，有错误请及时留言告知

bao文
关注
  • 0
    点赞
  • 踩
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
专栏目录
pointnet++自制室外城市区域数据集
08-04
《PointNet++自制室外城市区域数据集详解》 在计算机视觉和深度学习领域，三维点云数据处理是一项重要任务，而PointNet++作为一种强大的点云处理框架，为理解和解析复杂三维环境提供了有效工具。本文将深入探讨...
PointNet++ 源码有批注
08-01
PointNet++是深度学习领域中处理三维点云数据的一种创新性网络架构，由斯坦福大学的研究团队在2017年提出。它扩展了原始的PointNet（PointNet1），在处理非结构化3D数据时增加了对局部结构的感知能力。在本压缩包中...
Pointnet++
weixin_43156255的博客
01-04 1946
pointnet++
最全PointNet和PointNet++要点梳理总结
DAOCHI
11-12 3万+
一、基本简介 本篇博文主要是对 PointNet，PointNet++ 论文的要点进行梳理和总结。认真阅读本博文后，不仅能够深刻理解论文的核心算法思想，而且对模型训练数据的准备和模型的训练细节也能了然于胸。如果想阅读原论文，参考下面的链接资源，附有本人对原论文的中文翻译。 PointNet论文翻译，附链接。 PointNet++论文翻译，附链接。 二、PointNet 解决了什么问题？ a. 由于卷积网络在2D图像上的兴起，很多研究者开始将神经网络应用于3D点云数据。但是，大部分工作都是将3D
PyTorch复现PointNet——模型训练+可视化测试显示
beyond谚语的博客
07-09 1368
PointNet源码调试（模型训练+可视化测试显示）
【3D计算机视觉】从PointNet到PointNet++理论及pytorch代码
热门推荐
小执着的博客
03-28 7万+
从PointNet到PointNet++理论及代码详解1. 点云是什么1.1 三维数据的表现形式1.2 为什么使用点云1.3 点云上以往的相关工作2. PointNet2.1 基于点云的置换不变性2.1.1 由对称函数到PointNet(vanilla)2.1.2 理论证明2.2 基于点云的旋转不变性2.3 网络总体结构2.4 实验结果和网络的鲁棒性2.5 代码解析3. PointNet++3.1...
PointNet、PointNet++原理解析
记录老板的每次pua，分享领导每天的毒鸡汤...
05-22 7281
pointNet、pointNet++算法学习笔记
pointnet推理部署--onnxruntime框架
taifyang的博客
07-23 1315
注意，由于C++无法直接读取npy格式文件（可以依赖一些库），这里先使用python脚本将npy文件转换成txt文件。以分16类50部分，gpu版本为例。以分10类，gpu版本为例。以分13类，gpu版本为例。
PointNet解读
xsh_roy的博客
05-09 1万+
PointNet解决的问题： 如上图所示： 1.点云图像的分类（整片点云是什么物体） 2.点云图像的部件分割（整片点云所代表的物体能拆分的结构） 3.点云图像的语义分割（将三维点云环境中不同的物体用不同的颜色区分开） 论文中展示的输入输出效果： 1.部件分割的效果（左边是输入不完整的点云，右边是输入完整的点云） 2.语义分割的效果 论文中提到的点云的三个特性： 1.点的无序性：点云是无序的集合，点与点之间没有严格的顺序，如将两个点交换之后仍然表示同一个点云。 2.点之间的相互作.
PointNet代码解析
cocapop的博客
04-07 711
然后直接基于这个“2D图像”做卷积，第一个卷积核size是 [1,3] ，正好对应的就是“2D图像”的一行，也就是一个点（三维坐标），输出通道数是64，因此输出张量维度应该是 B×N×1×64。这里input_image维度是 B×N×3×1 ，因此将点云看成是W和H分为N和3的2D图像，维度是 1。第二个卷积核size是 [1,1] ， 1∗1 卷积只改变通道数，输出张量维度是 B×N×1×64。max_pooling，这个在论文的图中还是代码中都有体现，代码甚至直接用注释注明了。
PointNet介绍
leo0308的博客
03-25 1万+
论文：PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation 代码：https://github.com/charlesq34/pointnet 0 引言 PointNet是处理点云数据的深度学习模型，其地位堪比2D图像处理中的CNN网络， 后续的诸多点云数据处理的深度网络都有PointNet的影子。 1 点云数据介绍 既然PointNet处理的是点云数据， 那么首先需要先知道点云数据是长什么样子。 其实所谓点
PointNet和PointNet++的Pytorch实现
08-31
（1）发布语义分割的预训练模型，其中PointNet++可以达到53.5%的mIoU。 (2) 发布预训练模型用于分类和部分分割log/。 2021/03/20：更新分类代码，包括： (1) 添加训练ModelNet10数据集的代码。使用--num_category...
pointnet++的.so文件包
09-02
点云处理在计算机视觉领域中占据着重要的地位，而PointNet++是这一领域的经典之作，它在原始PointNet的基础上进行了扩展，增强了对局部结构的捕捉能力。本压缩包提供了PointNet++实现的一部分关键组件——`.so`文件...
Pointnet2_PyTorch:Pointnet2Pointnet++的PyTorch实现
08-03
Pointnet2/Pointnet++ PyTorch 项目状态：未维护。 由于时间有限，我没有更新此代码的计划，也不会响应问题。 用 PyTorch 编写的 Pointnet2/Pointnet++ 的。 通过支持多 GPU。 支持 PyTorch 版本 >= 1.0.0。 使用来...
ASP外观专利图像检索平台(源代码+论文).rar
07-21
ASP外观专利图像检索平台(源代码+论文)
C++课程设计：电煤气管理系统【源码+文档】
07-21
C++课程设计：电煤气管理系统【源码+文档】 本程序是一个水电气管理信息系统，能够对高校的水电气费用进行管理, 包括了成员的基本信息，如学号、编号、姓名、成员水电气的用量等。程序的用途包括缴纳水电气费、查询一个同学水电气费用量、查看所有同学的缴费情况、增加学生信息、删除学生信息、退出系统等。在设计时也考虑到学生和教师在用水电气时的不同，学生可以免费使用一定额度的水电气，超过这个额度的以后必须付费，且付费部分水电气费的价格要高于教工的收费标准，该措施的实行是为了鼓励同学们节约资源，以免造成不必要的资源浪费。该软件主要是为了学校的管理人员提供便捷，以更快的完成水电气费用的收缴。该软件本着简洁明了，实用稳定为一体进行设计。 系统将实现以下功能： （1）实现对用户信息的录入； （2）实现水电煤气数据的录入； （3）实现计算并查询用户应缴费用，查询未缴纳费用的名单； （4）实现对人员的删除和添加；
ASP+ACCESS网上动态同学录系统（源代码+论文+系统说明+答辩PPT）.rar
07-21
ASP+ACCESS网上动态同学录系统（源代码+论文+系统说明+答辩PPT）
avaloniassification-mas笔记
最新发布
07-21
avalonia
实现pointnet
05-22
PointNet是一个用于点云分类、分割和检测的深度学习网络，它是首个仅仅基于点云的端到端深度学习架构。PointNet的实现主要包括以下步骤： 1. 数据预处理：将点云转换为网络可以处理的形式，一般是将点云投影到一个...

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值