Skip to content

P
PaddleSlim

PaddlePaddle / PaddleSlim
1 年多 前同步成功

通知 51
Star 1434
Fork 344
P
PaddleSlim
切换分支/标签
查找文件 普通视图历史永久链接Permalink
index.html 14.6 KB
Newer Older
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
1 
<!DOCTYPE html>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
2 3 4 5 6 7 8 9 10 
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
  <meta charset="utf-8">
  <meta http-equiv="X-UA-Compatible" content="IE=edge">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  <meta name="description" content="None">
  
  <link rel="shortcut icon" href="img/favicon.ico">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
11 
  <title>Home - PaddleSlim Docs</title>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
12 
  <link href='https://fonts.googleapis.com/css?family=Lato:400,700|Roboto+Slab:400,700|Inconsolata:400,700' rel='stylesheet' type='text/css'>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
13
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
14 15 16 17 18 19 
  <link rel="stylesheet" href="css/theme.css" type="text/css" />
  <link rel="stylesheet" href="css/theme_extra.css" type="text/css" />
  <link rel="stylesheet" href="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/github.min.css">
  
  <script>
    // Current page data
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
20 
    var mkdocs_page_name = "Home";
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
21 22 23 24 25 26 27 28 29 30 
    var mkdocs_page_input_path = "index.md";
    var mkdocs_page_url = null;
  </script>
  
  <script src="js/jquery-2.1.1.min.js" defer></script>
  <script src="js/modernizr-2.8.3.min.js" defer></script>
  <script src="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/highlight.min.js"></script>
  <script>hljs.initHighlightingOnLoad();</script> 
  
</head>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
31
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
32 
<body class="wy-body-for-nav" role="document">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
33
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
34 
  <div class="wy-grid-for-nav">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
35
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
36 37 38 
    
    <nav data-toggle="wy-nav-shift" class="wy-nav-side stickynav">
      <div class="wy-side-nav-search">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
39 
        <a href="." class="icon icon-home"> PaddleSlim Docs</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
40 41 42 43 44 45 
        <div role="search">
  <form id ="rtd-search-form" class="wy-form" action="./search.html" method="get">
    <input type="text" name="q" placeholder="Search docs" title="Type search term here" />
  </form>
</div>
      </div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
46
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
47 48 49 50 51 52 
      <div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
	<ul class="current">
	  
          
            <li class="toctree-l1 current">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
53 
    <a class="current" href=".">Home</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
54 
    <ul class="subnav">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
55
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
56 
    <li class="toctree-l2"><a href="#paddleslim">PaddleSlim</a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
57 58 59 
    
        <ul>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
60 
            <li><a class="toctree-l3" href="#_1">架构</a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
61
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
62 
            <li><a class="toctree-l3" href="#_2">功能</a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
63
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
64 
            <li><a class="toctree-l3" href="#_3">安装</a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
65
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
66 67 68 
            <li><a class="toctree-l3" href="#_4">性能</a></li>
        
            <li><a class="toctree-l3" href="#_8">最佳实践</a></li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
69 70 71 72 
        
        </ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
73 
    </ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
74 75 76 77 
	    </li>
          
            <li class="toctree-l1">
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
78 79 80 81 82 
    <a class="" href="model_zoo/">模型库</a>
	    </li>
          
            <li class="toctree-l1">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
83 
    <span class="caption-text">教程</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
84 85 86 
    <ul class="subnav">
                <li class="">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
87 
    <a class="" href="tutorials/quant_post_demo/">离线量化</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
88 
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
89 
                <li class="">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
90
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
91 
    <a class="" href="tutorials/quant_aware_demo/">量化训练</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
92 
                </li>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
93 94 
                <li class="">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
95 
    <a class="" href="tutorials/quant_embedding_demo/">Embedding量化</a>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
96 
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
97 
                <li class="">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
98
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
99 
    <a class="" href="tutorials/nas_demo/">SA搜索</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
100 
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
101 
                <li class="">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
102
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
103 
    <a class="" href="tutorials/distillation_demo/">知识蒸馏</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
104 
                </li>
B
Deployed 9bcff00 with MkDocs version: 1.0.4  
baiyfbupt 已提交
105 106 107 108 109 
    </ul>
	    </li>
          
            <li class="toctree-l1">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
110 
    <span class="caption-text">API</span>
B
Deployed 9bcff00 with MkDocs version: 1.0.4  
baiyfbupt 已提交
111 
    <ul class="subnav">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
112 
                <li class="">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
113
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
114 
    <a class="" href="api/quantization_api/">量化</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
115 
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
116 
                <li class="">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
117
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
118 
    <a class="" href="api/prune_api/">剪枝与敏感度</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
119 
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
120 
                <li class="">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
121
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
122 
    <a class="" href="api/analysis_api/">模型分析</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
123 
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
124 125 
                <li class="">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
126 
    <a class="" href="api/single_distiller_api/">知识蒸馏</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
127 
                </li>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
128 129 
                <li class="">
B
Deployed 55210fe with MkDocs version: 1.0.4  
baiyfbupt 已提交
130 131 132 133 134 135 136 137 138 
    <a class="" href="api/nas_api/">SA搜索</a>
                </li>
                <li class="">
                    
    <a class="" href="api/search_space/">搜索空间</a>
                </li>
                <li class="">
                    
    <a class="" href="table_latency/">硬件延时评估表</a>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
139 
                </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
140 
    </ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
141 142 
	    </li>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
143 144 145 146 147 
            <li class="toctree-l1">
		
    <a class="" href="algo/algo/">算法原理</a>
	    </li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
148 149 150 151 
        </ul>
      </div>
      &nbsp;
    </nav>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
152
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
153 
    <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
154
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
155 156 157 
      
      <nav class="wy-nav-top" role="navigation" aria-label="top navigation">
        <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
158 
        <a href=".">PaddleSlim Docs</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
159 
      </nav>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
160
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
161 162 163 164 165 166 167 168 169 
      
      <div class="wy-nav-content">
        <div class="rst-content">
          <div role="navigation" aria-label="breadcrumbs navigation">
  <ul class="wy-breadcrumbs">
    <li><a href=".">Docs</a> &raquo;</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
170 
    <li>Home</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
171 172 
    <li class="wy-breadcrumbs-aside">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
173 174 175 
        <a href="https://github.com/PaddlePaddle/PaddleSlim/edit/master/docs/index.md"
          class="icon icon-github"> Edit on GitHub</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
176 177 178 179 180 181 182 
    </li>
  </ul>
  <hr/>
</div>
          <div role="main">
            <div class="section">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
183 
                <h1 id="paddleslim">PaddleSlim<a class="headerlink" href="#paddleslim" title="Permanent link">#</a></h1>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
184 
<p><a href="https://github.com/PaddlePaddle/PaddleSlim">PaddleSlim</a>是PaddlePaddle框架的一个子模块，主要用于压缩图像领域模型。在PaddleSlim中，不仅实现了目前主流的网络剪枝、量化、蒸馏三种压缩策略，还实现了超参数搜索和小模型网络结构搜索功能。在后续版本中，会添加更多的压缩策略，以及完善对NLP领域模型的支持。</p>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
185 186 187 
<h2 id="_1">架构<a class="headerlink" href="#_1" title="Permanent link">#</a></h2>
<p>(TODO) PaddleSlim架构介绍...</p>
<h2 id="_2">功能<a class="headerlink" href="#_2" title="Permanent link">#</a></h2>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
188 
<ul>
B
Deployed 76338d9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
189 
<li>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
190 191 192 193 194 195 
<p>模型剪裁</p>
<ul>
<li>通道均匀模型剪裁</li>
<li>基于敏感度的模型剪裁</li>
<li>基于进化算法的自动模型剪裁</li>
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
196 197 198 
</li>
<li>
<p>量化训练</p>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
199 200 201 202 203 
<ul>
<li>在线量化训练</li>
<li>离线量化</li>
<li>对权重全局量化和Channel-Wise量化</li>
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
204 205 
</li>
<li>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
206 
<p>知识蒸馏</p>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
207 208 209 
</li>
<li>
<p>轻量神经网络结构自动搜索（Light-NAS）</p>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
210 
<ul>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
211 
<li>基于进化算法的轻量神经网络结构自动搜索</li>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
212 
<li>支持FLOPS / 硬件延时约束</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
213 214 
<li>支持多平台模型延时评估</li>
</ul>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
215 216 217 
</li>
</ul>
<h2 id="_3">安装<a class="headerlink" href="#_3" title="Permanent link">#</a></h2>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
218 219 220 221 
<p>安装PaddleSlim前，请确认已正确安装Paddle1.6版本或更新版本。Paddle安装请参考：<a href="https://www.paddlepaddle.org.cn/install/quick">Paddle安装教程</a>。</p>
<ul>
<li>安装develop版本</li>
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
222 223 224 
<table class="codehilitetable"><tr><td class="linenos"><div class="linenodiv"><pre><span></span>1
2
3</pre></div></td><td class="code"><div class="codehilite"><pre><span></span><span class="n">git</span> <span class="n">clone</span> <span class="n">https</span><span class="p">:</span><span class="o">//</span><span class="n">github</span><span class="p">.</span><span class="n">com</span><span class="o">/</span><span class="n">PaddlePaddle</span><span class="o">/</span><span class="n">PaddleSlim</span><span class="p">.</span><span class="n">git</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
225 226 227 
<span class="n">cd</span> <span class="n">PaddleSlim</span>
<span class="n">python</span> <span class="n">setup</span><span class="p">.</span><span class="n">py</span> <span class="n">install</span>
</pre></div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
228 
</td></tr></table>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
229 230 231 232 

<ul>
<li>安装官方发布的最新版本</li>
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
233 
<table class="codehilitetable"><tr><td class="linenos"><div class="linenodiv"><pre><span></span>1</pre></div></td><td class="code"><div class="codehilite"><pre><span></span><span class="n">pip</span> <span class="n">install</span> <span class="n">paddleslim</span> <span class="o">-</span><span class="n">i</span> <span class="n">https</span><span class="p">:</span><span class="o">//</span><span class="n">pypi</span><span class="p">.</span><span class="n">org</span><span class="o">/</span><span class="k">simple</span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
234 
</pre></div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
235 
</td></tr></table>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
236 237 238 239 240 

<ul>
<li>安装历史版本</li>
</ul>
<p>请点击<a href="https://pypi.org/project/paddleslim/#history">pypi.org</a>查看可安装历史版本。</p>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 
<h2 id="_4">性能<a class="headerlink" href="#_4" title="Permanent link">#</a></h2>
<p>PaddleSlim的离线量化、在线量化、均匀剪裁、敏感度剪裁、知识蒸馏等策略在多个主流CNN模型上都进行了验证，部分实验结果如下，全部的实验结果请参见<a href="">模型库</a></p>
<h4 id="_5">量化<a class="headerlink" href="#_5" title="Permanent link">#</a></h4>
<p>我们在多个模型上分别验证了离线量化和在线量化的效果：离线量化过程简单，无需再次训练，但精度损失相对较多；在线量化需要额外的训练过程，但好处是精度损失可以被有效降低：</p>
<table>
<thead>
<tr>
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">Top-1/Top-5 Acc</th>
<th align="center">模型大小（MB）</th>
</tr>
</thead>
<tbody>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">-</td>
<td align="center">72.15%/90.65%</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">quant_post</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">quant_aware</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">ResNet50</td>
<td align="center">-</td>
<td align="center">76.50%/93.00%</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">ResNet50</td>
<td align="center">quant_post</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">ResNet50</td>
<td align="center">quant_aware</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
</tr>
</tbody>
</table>
<h4 id="_6">剪裁<a class="headerlink" href="#_6" title="Permanent link">#</a></h4>
<p>模型剪裁在不带来过多精度损失的前提下有效降低模型计算的FLOPs（floating point operations per second），PaddleSlim在多个模型上验证了均匀剪裁、敏感度剪裁、自动剪裁三种方式，并取得了一定收益：</p>
<table>
<thead>
<tr>
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">Top-1/Top-5 Acc</th>
<th align="center">模型大小（MB）</th>
<th align="center">FLOPs（M）</th>
</tr>
</thead>
<tbody>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">-</td>
<td align="center">72.15%/90.65%</td>
<td align="center">xx</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">uniform -xx%</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">sensitive -xx%</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">ResNet34</td>
<td align="center">-</td>
<td align="center">74.57%/92.14%</td>
<td align="center">xx</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">ResNet34</td>
<td align="center">uniform -xx%</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center">xx</td>
</tr>
<tr>
<td align="center">ResNet34</td>
<td align="center">auto -xx%</td>
<td align="center">xx%/xx%</td>
<td align="center">xx</td>
<td align="center">xx</td>
</tr>
</tbody>
</table>
<h4 id="_7">蒸馏<a class="headerlink" href="#_7" title="Permanent link">#</a></h4>
<p>我们在多个图象分类和目标检测模型上应用了PaddleSlim的知识蒸馏功能，这些实验在不增加目标模型计算复杂度的前提下都一定程度提升了其精度表现：</p>
<table>
<thead>
<tr>
<th align="center">模型</th>
<th align="center">压缩方法</th>
<th align="center">Top-1/Top-5 Acc</th>
</tr>
</thead>
<tbody>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">-</td>
<td align="center">72.15%/90.65%</td>
</tr>
<tr>
<td align="center">MobileNetV2</td>
<td align="center">ResNet50_vd distill</td>
<td align="center">74.28%/91.53%</td>
</tr>
<tr>
<td align="center">ResNet50</td>
<td align="center">-</td>
<td align="center">76.50%/93.00%</td>
</tr>
<tr>
<td align="center">ResNet50</td>
<td align="center">ResNet101 distill</td>
<td align="center">77.29%/93.65%</td>
</tr>
</tbody>
</table>
<h2 id="_8">最佳实践<a class="headerlink" href="#_8" title="Permanent link">#</a></h2>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
384 
<ul>
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
385 386 
<li><a href="">Paddle检测库</a>：介绍如何在PaddleDetection检测库中使用PaddleSlim。</li>
<li><a href="">Paddle分割库</a>：介绍如何在PaddleSeg分割库中使用PaddleSlim。</li>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
387 388 
<li><a href="">PaddleLite</a>：介绍如何使用预测库PaddleLite部署PaddleSlim产出的模型。</li>
</ul>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
389
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
390 
            </div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
391 392 393 394 395 
          </div>
          <footer>
  
    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
396 
        <a href="model_zoo/" class="btn btn-neutral float-right" title="模型库">Next <span class="icon icon-circle-arrow-right"></span></a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
397 398
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
399 
    </div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
400 401 402 403 404 405 406 407 408 409 410 411 
  

  <hr/>

  <div role="contentinfo">
    <!-- Copyright etc -->
    
  </div>

  Built with <a href="http://www.mkdocs.org">MkDocs</a> using a <a href="https://github.com/snide/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</footer>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
412 
        </div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
413 414 415 416 417 418 419 420 421 
      </div>

    </section>

  </div>

  <div class="rst-versions" role="note" style="cursor: pointer">
    <span class="rst-current-version" data-toggle="rst-current-version">
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
422 423 
          <a href="https://github.com/PaddlePaddle/PaddleSlim/" class="fa fa-github" style="float: left; color: #fcfcfc"> GitHub</a>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
424 425
B
Deployed 482d368 with MkDocs version: 1.0.4  
baiyfbupt 已提交
426 
        <span style="margin-left: 15px"><a href="model_zoo/" style="color: #fcfcfc">Next &raquo;</a></span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
427 428 
      
    </span>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
429 
</div>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
430 431 
    <script>var base_url = '.';</script>
    <script src="js/theme.js" defer></script>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
432 
      <script src="mathjax-config.js" defer></script>
B
Deployed 0f9c3ce with MkDocs version: 1.0.4  
baiyfbupt 已提交
433 
      <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS-MML_HTMLorMML" defer></script>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
434 
      <script src="search/main.js" defer></script>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
435
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
436 
</body>
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
437 438 439 440 
</html>

<!--
MkDocs version : 1.0.4
B
Deployed 0b29482 with MkDocs version: 1.0.4  
baiyfbupt 已提交
441 
Build Date UTC : 2020-01-08 10:42:25
B
Deployed 09f99c9 with MkDocs version: 1.0.4  
baiyfbupt 已提交
442 
-->