天天看點

MobileNet v11. 深度可分離卷積

https://www.cnblogs.com/darkknightzh/p/9410540.html

<div id="post_detail">
<div class="post">
	<h2>
		<a id="cb_post_title_url" href="https://www.cnblogs.com/darkknightzh/p/9410540.html" target="_blank" rel="external nofollow"  target="_blank" rel="external nofollow"  target="_blank" rel="external nofollow" >(原)MobileNetV1</a>
	</h2>
	<div class="postbody">
	<div id="cnblogs_post_body" class="blogpost-body"><a name="_labelTop"></a><div id="navCategory"><p style="font-size:16px"><b><span style="color: #d2681e;">~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~</span></b></p><p style="font-size:24px"><b><strong><span style="color: #d2681e;">本文目錄</span></strong></b></p><ul class="first_class_ul"><li><a href="#_label0" target="_blank" rel="external nofollow" >1. 深度可分離卷積</a></li><li><a href="#_label1" target="_blank" rel="external nofollow" >2. 網絡結構</a></li><li><a href="#_label2" target="_blank" rel="external nofollow" >3. 寬度縮放因子(width multiplier)</a></li><li><a href="#_label3" target="_blank" rel="external nofollow" >4. 分辨率縮放因子(resolution multiplier)</a></li><li><a href="#_label4" target="_blank" rel="external nofollow" >5. pytorch代碼</a></li></ul></div><p style="font-size:16px"><b><span style="color: #d2681e;">~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~</span></b></p><p>轉載請注明出處:</p>
           

https://www.cnblogs.com/darkknightzh/p/9410540.html

論文:

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

網址:

https://arxiv.org/abs/1704.04861?context=cs

非官方的pytorch代碼:

https://github.com/marvis/pytorch-mobilenet

回到頂部(go to top)

1. 深度可分離卷積

mobilenetV1使用的是深度可分離卷積(Depthwise Separable Convolution,DSC),DSC包含兩部分:depthwise convolution(DWC)+ pointwise convolution(PWC)。DWC對輸入的通道進行濾波,其不增加通道的數量,PWC用于将PWC不同的通道進行連接配接,其可以增加通道的數量。通過這種分解的方式,可以明顯的減少計算量。

如下圖所示,傳統的卷積(a),卷積核參數為DK⋅DK⋅M⋅NDK⋅DK⋅M⋅N

« 上一篇: (原)Show, Attend and Translate: Unsupervised Image Translation with Self-Regularization and Attention

» 下一篇: (原)MobileNetV2

</div>
	<p class="postfoot">
		posted on <span id="post-date">2018-08-02 22:12</span> <a href="https://www.cnblogs.com/darkknightzh/" target="_blank" rel="external nofollow" >darkknightzh</a> 閱讀(<span id="post_view_count">2750</span>) 評論(<span id="post_comment_count">0</span>)  <a href="https://i.cnblogs.com/EditPosts.aspx?postid=9410540" target="_blank" rel="external nofollow"  rel="nofollow">編輯</a> <a href="#" target="_blank" rel="external nofollow"  target="_blank" rel="external nofollow"  onclick="AddToWz(9410540);return false;">收藏</a>
	</p>
</div>
<script type="text/javascript">var allowComments=true,cb_blogId=201467,cb_entryId=9410540,cb_blogApp=currentBlogApp,cb_blogUserGuid='2ff23002-a055-e411-b908-9dcfd8948a71',cb_entryCreatedDate='2018/8/2 22:12:00';loadViewCount(cb_entryId);var cb_postType=1;var isMarkdown=false;</script>

</div><a name="!comments"></a><div id="blog-comments-placeholder"></div><script type="text/javascript">var commentManager = new blogCommentManager();commentManager.renderComments(0);</script>
           

重新整理評論 重新整理頁面 傳回頂部 注冊使用者登入後才能發表評論,請 登入 或 注冊, 通路網站首頁。 【推薦】超50萬C++/C#源碼: 大型實時仿真組态圖形源碼

【前端】SpreadJS表格控件,可嵌入系統開發的線上Excel

【推薦】程式員問答平台,解決您開發中遇到的技術難題

繼續閱讀