Caffe for Python 官方教程(翻译)

导言

       本教程中，我们将会利用Caffe官方提供的深度模型——CaffeNet（该模型是基于Krizhevsky等人的模型的）来演示图像识别与分类。我们将分别用CPU和GPU来进行演示，并对比其性能。然后深入探讨该模型的一些其它特征。

1、准备工作

       1.1 首先，安装Python，numpy以及matplotlib。

<code class="hljs python has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"> <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#安装Python环境、numpy、matplotlib</span>
<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">import</span> numpy <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">as</span> np
<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">import</span> matplotlib.pyplot <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">as</span> plt
%matplotlib inline

<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#设置默认显示参数</span>
plt.rcParams[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'figure.figsize'</span>] = (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">10</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">10</span>)        <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 图像显示大小</span>
plt.rcParams[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'image.interpolation'</span>] = <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'nearest'</span>  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 最近邻差值: 像素为正方形</span>
plt.rcParams[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'image.cmap'</span>] = <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'gray'</span>  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 使用灰度输出而不是彩色输出</span></code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li></ul>

       1.2 然后，加载Load caffe。

# caffe模块要在Python的路径下;
# 这里我们将把caffe 模块添加到Python路径下.
import sys
caffe_root = '../'  #该文件要从路径{caffe_root}/examples下运行，否则要调整这一行。
sys.path.insert(0, caffe_root + 'python')

import caffe
# 如果你看到"No module named _caffe",那么要么就是你没有正确编译pycaffe；要么就是你的路径有错误。

说明：该步骤，本人是将编译好的pycaffe文件下的全部东西复制到Python的“site-packages”下的。所以不知道按上述做法具体会出现什么问题。

       1.3 必要的话，需要事先下载“CaffeNet”模型，该模型是AlexNet的变形。

<code class="hljs lua has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">import <span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">os</span>
<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">if</span> <span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">os</span>.path.isfile(caffe_root + <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel'</span>):
<span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">print</span> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'CaffeNet found.'</span>
<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">else</span>:
<span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">print</span> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'Downloading pre-trained CaffeNet model...'</span>
!../scripts/download_model_binary.py ../models/bvlc_reference_caffenet</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li></ul>

说明：该步骤，本人是事先下载好”bvlc_reference_caffenet.caffemodel”，然后将其放在”caffe_root + ‘models/bvlc_reference_caffenet/”目录下面，因为用代码下载太慢了。

2、加载网络并设置输入预处理

       2.1 将Caffe设置为CPU模式，并从硬盘加载网络。

<code class="hljs bash has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">
caffe.set_mode_cpu()

model_def = caffe_root + <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'models/bvlc_reference_caffenet/deploy.prototxt'</span>
model_weights = caffe_root + <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel'</span>

net = caffe.Net(model_def,      <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 定义模型结构</span>
model_weights,  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 包含了模型的训练权值</span>
caffe.TEST)     <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 使用测试模式(不执行dropout)</span></code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li></ul>

       2.2 设置输入预处理(我们将用Caffe’s caffe.io.Transformer来进行预处理。不过该步骤与caffe的其它模块是相互独立的，所以任何预处理代码应该都是可行的)。我们使用的CaffeNet模型默认的输入图像格式是BGR格式的，其像素值位于[0,255]之间，同时每个像素值都减去了ImageNet图像的平均值。除此之外，通道的维数等于第一维(outermost)的大小。另外，因为matplotlib加载的图像的值位于[0,1]之间，并且格式是RGB格式，通道的维数等于innermost的维数，所以我们需要做一些变换(感觉这一段翻译的太烂)，如下：

<code class="hljs php has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">
<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 加载ImageNet图像均值 (随着Caffe一起发布的)</span>
mu = np.load(caffe_root + <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'python/caffe/imagenet/ilsvrc_2012_mean.npy'</span>)
mu = mu.mean(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>).mean(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>)  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#对所有像素值取平均以此获取BGR的均值像素值</span>
<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">print</span> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'mean-subtracted values:'</span>, zip(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'BGR'</span>, mu)

<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 对输入数据进行变换</span>
transformer = caffe.io.Transformer({<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>: net.blobs[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>].data.shape})

transformer.set_transpose(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>, (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">2</span>,<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>,<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>))  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#将图像的通道数设置为outermost的维数</span>
transformer.set_mean(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>, mu)            <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#对于每个通道，都减去BGR的均值像素值</span>
transformer.set_raw_scale(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">255</span>)      <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#将像素值从[0,255]变换到[0,1]之间</span>
transformer.set_channel_swap(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>, (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">2</span>,<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>,<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>))  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#交换通道，从RGB变换到BGR</span></code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li></ul>

3、用CPU分类

       3.1 现在我们开始进行分类。尽管我们只对一张图像进行分类，不过我们将batch的大小设置为50以此来演示batching。

<code class="hljs bash has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 设置输入图像大小</span>
net.blobs[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>].reshape(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>,        <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># batch 大小</span>
<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>,         <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 3-channel (BGR) images</span>
<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">227</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">227</span>)  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 图像大小为:227x227</span></code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul>

       3.2 加载图像(caffe自带的)并进行预处理。

<code class="hljs mel has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">    <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">image</span> = caffe.io.load_image(caffe_root + <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'examples/images/cat.jpg'</span>)
transformed_image = transformer.preprocess(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>, <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">image</span>)
plt.imshow(<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">image</span>)
plt.show()</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul>

说明：这里的”plt.show()”是我自己加的，不加的话没法显示图像。

       3.3 接下来，开始进行识别分类

<code class="hljs r has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 将图像数据拷贝到为net分配的内存中</span>
net.blobs[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>].data[<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">...</span>] = transformed_image

<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">### 执行分类</span>
output = net.forward()
output_prob = output[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'prob'</span>][<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>]  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;">#batch中第一张图像的概率值   </span>
print <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'predicted class is:'</span>, output_prob.argmax()</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li></ul>

predicted class is: 281

       网络输出是一个概率向量；最可能的类别是第281个类别。但是结果是否正确呢，让我们来查看一下ImageNet的标签。

<code class="hljs perl has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">    <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 加载ImageNet标签</span>
labels_file = caffe_root + <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data/ilsvrc12/synset_words.txt'</span>
<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">if</span> <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">not</span> os.path.<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">exists</span>(labels_file):
!../data/ilsvrc12/get_ilsvrc_aux.sh

labels = np.loadtxt(labels_file, str, delimiter=<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'\t'</span>)

<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">print</span> <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'output label:'</span>, labels[output_prob.argmax()]</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li></ul>

说明：ImageNet标签文件(synset_words.txt)需要自己下载

output label: n02123045 tabby, tabby cat

       ”Tabby cat”是正确的，然后我们再来看下其它几个置信的较高的结果。

<code class="hljs oxygene has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"># sort top five predictions <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">from</span> softmax output
top_inds = output_prob.argsort()[::-<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>][:<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">5</span>]  # <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">reverse</span> sort <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">and</span> <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">take</span> five largest items

print <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'probabilities and labels:'</span>
zip(output_prob[top_inds], labels[top_inds])</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li></ul>

probabilities and labels:

<code class="hljs json has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">[(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0.31243637</span>, 'n02123045 tabby, tabby cat'),
(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0.2379719</span>, 'n02123159 tiger cat'),
(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0.12387239</span>, 'n02124075 Egyptian cat'),
(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0.10075711</span>, 'n02119022 red fox, Vulpes vulpes'),
(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0.070957087</span>, 'n02127052 lynx, catamount')]</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li></ul>

我们可以看出，较低置信度的结构也是合理的。

4、GPU模式

       4.1 让我们先看下CPU的分类时间，然后再与GPU进行比较。

<code class="hljs haml has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-tag" style="color: rgb(0, 102, 102); box-sizing: border-box;">    %<span class="hljs-title" style="box-sizing: border-box; color: rgb(0, 0, 136);">timeit</span></span> net.forward()</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li></ul>

1 loop, best of 3: 1.42 s per loop

还是需要一段时间的，即使是对批量的50张图像。然后，让我们看下GPU模式下的运行时间。

<code class="hljs avrasm has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">caffe<span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.set</span>_device(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>)  <span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;"># 如果你有多个GPU，那么选择第一个</span>
caffe<span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.set</span>_mode_gpu()
net<span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.forward</span>()  <span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;"># run once before timing to set up memory</span>
%timeit net<span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.forward</span>()</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul>

10 loops, best of 3: 70.2 ms per loop

这下就快多了。

5、测试网络的中间层输出

       我们的网络不单单是一个黑盒子。接下来，我们来看下该模型的一些参数和一些中间输出。首先，我们来看下如何读取网络的结构(每层的名字以及相应层的参数)。对于每一层，其结构构成为:(batch_size, channel_dim, height, width)。

<code class="hljs haskell has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;"># 对于每一层，显示输出类型。</span>
for layer_name, blob <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">in</span> net.blobs.iteritems():
print layer_name + '\t' + str(blob.<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.shape)</span>

<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>    <span class="hljs-container" style="box-sizing: border-box;">(50, 3, 227, 227)</span></span>
conv1   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">96</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">55</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">55</span>)
pool1   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">96</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">27</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">27</span>)
norm1   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">96</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">27</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">27</span>)
conv2   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">27</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">27</span>)
pool2   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>)
norm2   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>)
conv3   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">384</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>)
conv4   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">384</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>)
conv5   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">13</span>)
pool5   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">6</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">6</span>)
fc6 (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>)
fc7 (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>)
fc8 (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1000</span>)
prob    (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">50</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1000</span>)</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li></ul>

       现在，我们来看下参数的形状。参数是OrderdDict类型，net.params。我们根据索引来访问参数。[0]:表示weights,[1]:表示biases。

参数形状的构成为:

(output_channels, input_channels, filter_height, filter_width)为weights;(output_channels,)为biases。

<code class="hljs avrasm has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">for layer_name, param <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">in</span> net<span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.params</span><span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.iteritems</span>():
print layer_name + <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'\t'</span> + str(param[<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>]<span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.data</span><span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.shape</span>), str(param[<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>]<span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.data</span><span class="hljs-preprocessor" style="color: rgb(68, 68, 68); box-sizing: border-box;">.shape</span>)

conv1   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">96</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">11</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">11</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">96</span>,)
conv2   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">48</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">5</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">5</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>,)
conv3   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">384</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">384</span>,)
conv4   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">384</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">192</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">384</span>,)
conv5   (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">192</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">256</span>,)
fc6 (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">9216</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>,)
fc7 (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>,)
fc8 (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1000</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">4096</span>) (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1000</span>,)</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li></ul>

       因为我们处理的是四位数据，所以我们将定义一个帮助函数来可视化特征。

<code class="hljs haskell has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-title" style="box-sizing: border-box;">def</span> vis_square(<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>):</span>
<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">""</span><span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"输入一个形如：(n, height, width) or (n, height, width, 3)的数组，并对每一个形如(height,width)的特征进行可视化sqrt(n) by sqrt(n)"</span><span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">""</span>

# 正则化数据
<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span> = <span class="hljs-container" style="box-sizing: border-box;">(<span class="hljs-title" style="box-sizing: border-box;">data</span> - <span class="hljs-title" style="box-sizing: border-box;">data</span>.<span class="hljs-title" style="box-sizing: border-box;">min</span>()</span>) / <span class="hljs-container" style="box-sizing: border-box;">(<span class="hljs-title" style="box-sizing: border-box;">data</span>.<span class="hljs-title" style="box-sizing: border-box;">max</span>()</span> - <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.min<span class="hljs-container" style="box-sizing: border-box;">()</span>)</span>

# 将滤波器的核转变为正方形
n = int(np.ceil(np.sqrt(<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.shape[0])))</span>
padding = (((<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>, n ** <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">2</span> - <span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.shape[0]),</span>
(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>), (<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>))                 # 在相邻的滤波器之间加入空白
+ ((<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>),) * (<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.ndim - 3))  # 不扩展最后一维</span>
<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span> = np.pad<span class="hljs-container" style="box-sizing: border-box;">(<span class="hljs-title" style="box-sizing: border-box;">data</span>, <span class="hljs-title" style="box-sizing: border-box;">padding</span>, <span class="hljs-title" style="box-sizing: border-box;">mode</span>='<span class="hljs-title" style="box-sizing: border-box;">constant'</span>, <span class="hljs-title" style="box-sizing: border-box;">constant_values</span>=1)</span>  # 扩展一个像素<span class="hljs-container" style="box-sizing: border-box;">(白色)</span></span>

# tile the filters into an image
<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span> = <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.reshape<span class="hljs-container" style="box-sizing: border-box;">((<span class="hljs-title" style="box-sizing: border-box;">n</span>, <span class="hljs-title" style="box-sizing: border-box;">n</span>)</span> + <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.shape[1:]).transpose<span class="hljs-container" style="box-sizing: border-box;">((0, 2, 1, 3)</span> + tuple<span class="hljs-container" style="box-sizing: border-box;">(<span class="hljs-title" style="box-sizing: border-box;">range</span>(4, <span class="hljs-title" style="box-sizing: border-box;">data</span>.<span class="hljs-title" style="box-sizing: border-box;">ndim</span> + 1)</span>))</span>
<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span> = <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.reshape<span class="hljs-container" style="box-sizing: border-box;">((<span class="hljs-title" style="box-sizing: border-box;">n</span> * <span class="hljs-title" style="box-sizing: border-box;">data</span>.<span class="hljs-title" style="box-sizing: border-box;">shape</span>[1], <span class="hljs-title" style="box-sizing: border-box;">n</span> * <span class="hljs-title" style="box-sizing: border-box;">data</span>.<span class="hljs-title" style="box-sizing: border-box;">shape</span>[3])</span> + <span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>.shape[4:])</span>

plt.imshow(<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>)</span>
plt.axis('off')
plt.show()</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li></ul>

       首先，我们来看下第一个卷积层(conv1)的输出特征。

<code class="hljs livecodeserver has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">    <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 参数为一个[weights, biases]的列表</span>
filters = net.<span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">params</span>[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'conv1'</span>][<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>].data
vis_square(filters.<span class="hljs-built_in" style="color: rgb(102, 0, 102); box-sizing: border-box;">transpose</span>(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">2</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>))</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li></ul>

上图为conv1的输出。

<code class="hljs haskell has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">    feat = net.blobs['conv1'].<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>[0, :36]</span>
vis_square(feat)</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li></ul>

上图为pool5的输出。

<code class="hljs haskell has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">    feat = net.blobs['pool5'].<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>[0]</span>
vis_square(feat)</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li></ul>

上图为第一个全连接层(fc6)的输出。

接下来，我们将显示输出结果及直方图。

<code class="hljs glsl has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">    feat = net.blobs['fc6'].data[<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>]
plt.subplot(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">2</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>)
plt.plot(feat.<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">flat</span>)
plt.subplot(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">2</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">2</span>)
_ = plt.hist(feat.<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">flat</span>[feat.<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">flat</span> > <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>], bins=<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">100</span>)
plt.show()</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li></ul>

上图为最终的概率输出，prob。

<code class="hljs haskell has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;">    feat = net.blobs['prob'].<span class="hljs-typedef" style="box-sizing: border-box;"><span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">data</span>[0]</span>
plt.figure(figsize=(<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">15</span>, <span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">3</span>))
plt.plot(feat.flat)
plt.show()</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li></ul>

上图显示了分类的聚类结果，峰值对应的标签为预测结果。

6、测试自己的图像

现在，我们随便从网上找一种图像，然后安装上述步骤来进行分类。

将”my_image_url”设为图像的链接（URL）

<code class="hljs r has-numbering" style="display: block; padding: 0px; color: inherit; box-sizing: border-box; font-family: 'Source Code Pro', monospace;font-size:undefined; white-space: pre; border-radius: 0px; word-wrap: normal; background: transparent;"><span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 下载图像</span>
my_image_url = <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">"..."</span>  <span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 将你的图像URL粘贴到这里</span>
<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 例如:</span>
<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># my_image_url = "https://upload.wikimedia.org/wikipedia/commons/b/be/Orang_Utan%2C_Semenggok_Forest_Reserve%2C_Sarawak%2C_Borneo%2C_Malaysia.JPG"</span>
!wget -O image.jpg $my_image_url

<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 变换图像并将其拷贝到网络</span>
image = caffe.io.load_image(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'image.jpg'</span>)
net.blobs[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>].data[<span class="hljs-keyword" style="color: rgb(0, 0, 136); box-sizing: border-box;">...</span>] = transformer.preprocess(<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'data'</span>, image)

<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 预测分类结果</span>
net.forward()

<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 获取输出概率值</span>
output_prob = net.blobs[<span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'prob'</span>].data[<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">0</span>]

<span class="hljs-comment" style="color: rgb(136, 0, 0); box-sizing: border-box;"># 将softmax的输出结果按照从大到小排序，并提取前5名</span>
top_inds = output_prob.argsort()[::-<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">1</span>][:<span class="hljs-number" style="color: rgb(0, 102, 102); box-sizing: border-box;">5</span>]

plt.imshow(image)
plt.show()
print <span class="hljs-string" style="color: rgb(0, 136, 0); box-sizing: border-box;">'probabilities and labels:'</span>
zip(output_prob[top_inds], labels[top_inds])</code><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; opacity: 0.098775; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li><li style="box-sizing: border-box; padding: 0px 5px;">21</li><li style="box-sizing: border-box; padding: 0px 5px;">22</li><li style="box-sizing: border-box; padding: 0px 5px;">23</li></ul><div class="save_code tracking-ad" data-mod="popu_249" style="box-sizing: border-box; position: absolute; height: 60px; right: 30px; top: 5px; color: rgb(255, 255, 255); cursor: pointer; z-index: 2;"><a target=_blank target="_blank" style="color: rgb(51, 102, 153); box-sizing: border-box;"><img src="http://static.blog.csdn.net/images/save_snippets.png" style="border: none; box-sizing: border-box; max-width: 100%;" alt="" /></a></div><ul class="pre-numbering" style="box-sizing: border-box; position: absolute; width: 50px; top: 0px; left: 0px; margin: 0px; padding: 6px 0px 40px; border-right-width: 1px; border-right-style: solid; border-right-color: rgb(221, 221, 221); list-style: none; text-align: right; background-color: rgb(238, 238, 238);"><li style="box-sizing: border-box; padding: 0px 5px;">1</li><li style="box-sizing: border-box; padding: 0px 5px;">2</li><li style="box-sizing: border-box; padding: 0px 5px;">3</li><li style="box-sizing: border-box; padding: 0px 5px;">4</li><li style="box-sizing: border-box; padding: 0px 5px;">5</li><li style="box-sizing: border-box; padding: 0px 5px;">6</li><li style="box-sizing: border-box; padding: 0px 5px;">7</li><li style="box-sizing: border-box; padding: 0px 5px;">8</li><li style="box-sizing: border-box; padding: 0px 5px;">9</li><li style="box-sizing: border-box; padding: 0px 5px;">10</li><li style="box-sizing: border-box; padding: 0px 5px;">11</li><li style="box-sizing: border-box; padding: 0px 5px;">12</li><li style="box-sizing: border-box; padding: 0px 5px;">13</li><li style="box-sizing: border-box; padding: 0px 5px;">14</li><li style="box-sizing: border-box; padding: 0px 5px;">15</li><li style="box-sizing: border-box; padding: 0px 5px;">16</li><li style="box-sizing: border-box; padding: 0px 5px;">17</li><li style="box-sizing: border-box; padding: 0px 5px;">18</li><li style="box-sizing: border-box; padding: 0px 5px;">19</li><li style="box-sizing: border-box; padding: 0px 5px;">20</li><li style="box-sizing: border-box; padding: 0px 5px;">21</li><li style="box-sizing: border-box; padding: 0px 5px;">22</li><li style="box-sizing: border-box; padding: 0px 5px;">23</li></ul>

实际中我对上述代码稍作了些改动代码：CaffePython。

点击查看原文