https://111qqz.com/en/series/caffe%E6%BA%90%E7%A0%81%E9%98%85%E8%AF%BB%E7%AC%94%E8%AE%B0/Recent content in Caffe源码阅读笔记 on 111qqz's blogHugo -- gohugo.ioenhust.111qqz@gmail.com (111qqz)hust.111qqz@gmail.com (111qqz)© 2011-2026 111qqzSat, 06 Aug 2022 12:26:03 +0800https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-06-30-caffe-source-code-analysis/Tue, 30 Jun 2020 19:47:02 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-06-30-caffe-source-code-analysis/<p>caffe做部署是YYDS!</p> <ol> <li> <p><a href="https://111qqz.github.io/2020/01/caffe-source-code-analysis-part1/" target="_blank" rel="noreferrer">blob</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/01/caffe-source-code-analysis-part2/" target="_blank" rel="noreferrer">layer</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/01/caffe-source-code-analysis-part3/" target="_blank" rel="noreferrer">net</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/04/caffe-source-code-analysis-part4/" target="_blank" rel="noreferrer">激活函数</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/04/caffe-source-code-analysis-part5/" target="_blank" rel="noreferrer">卷积</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/04/caffe-source-code-analysis-part6/" target="_blank" rel="noreferrer">reshape</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/04/caffe-source-code-analysis-part7/" target="_blank" rel="noreferrer">slice</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/04/caffe-source-code-analysis-part8/" target="_blank" rel="noreferrer">loss function</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/05/caffe-source-code-analysis-part9/" target="_blank" rel="noreferrer">reduce</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/05/caffe-source-code-analysis-part10/" target="_blank" rel="noreferrer">eltwise</a></p> </li> <li> <p><a href="https://111qqz.github.io/2020/05/caffe-source-code-analysis-part11/" target="_blank" rel="noreferrer">argmax</a></p> </li> </ol>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2022-08-06-caffe-source-code-analysis_part12/Sat, 06 Aug 2022 12:26:03 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2022-08-06-caffe-source-code-analysis_part12/<h2 class="relative group">背景 <div id="背景" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%83%8c%e6%99%af" aria-label="Anchor">#</a> </span> </h2> <p>2022年惊讶的发现，当时竟然没有写关于softmax的笔记，因此来补充一下。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-05-06-caffe-source-code-analysis_part11/Wed, 06 May 2020 21:26:03 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-05-06-caffe-source-code-analysis_part11/<h2 class="relative group">背景 <div id="背景" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%83%8c%e6%99%af" aria-label="Anchor">#</a> </span> </h2> <p>似乎没什么背景，继续看 caffe 代码。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-05-03-caffe-source-code-analysis_part10/Sun, 03 May 2020 17:53:22 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-05-03-caffe-source-code-analysis_part10/<h2 class="relative group">背景 <div id="背景" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%83%8c%e6%99%af" aria-label="Anchor">#</a> </span> </h2> <p>这个 layer 和 reduce layer 有一些相似，就干脆一起看了。 作用是输入至少两个 blob，然后对每个 blob 中的元素做一些运算，最后得到一个 blob。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-05-03-caffe-source-code-analysis_part9/Sun, 03 May 2020 15:16:53 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-05-03-caffe-source-code-analysis_part9/<h2 class="relative group">背景 <div id="背景" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%83%8c%e6%99%af" aria-label="Anchor">#</a> </span> </h2> <p>其实没什么背景，继续啃 caffe 代码而已 2333。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-18-caffe-source-code-analysis_part8/Sat, 18 Apr 2020 18:33:29 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-18-caffe-source-code-analysis_part8/<h2 class="relative group">背景 <div id="背景" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%83%8c%e6%99%af" aria-label="Anchor">#</a> </span> </h2> <p>虽然不太care 训练的过程，<del>但是由于容易看懂的layer都看得差不多了</del> 所以打算看一下这些loss function.</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-13-caffe-source-code-analysis_part7/Mon, 13 Apr 2020 21:22:54 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-13-caffe-source-code-analysis_part7/<h2 class="relative group">背景　 <div id="背景" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%83%8c%e6%99%af" aria-label="Anchor">#</a> </span> </h2> <p>OCR 组那边有个 shufflenet 的网络，里面有个 pytorch op 叫 chunk，转成的 onnx 对应的 op 是 <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Split" target="_blank" rel="noreferrer">split</a>。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-09-caffe-source-code-analysis_part6/Thu, 09 Apr 2020 21:03:06 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-09-caffe-source-code-analysis_part6/<h2 class="relative group">背景　 <div id="背景" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%83%8c%e6%99%af" aria-label="Anchor">#</a> </span> </h2> <p>最近在魔改 TensorRT 的 caffe parser。 之前 caffe 模型转到 trt 模型时，有一个修改需要将 reshape layer 的 param 末尾补 1，比较繁琐，于是看了下 caffe 的 reshape layer 的实现。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-08-caffe-source-code-analysis_part5/Wed, 08 Apr 2020 20:29:37 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-08-caffe-source-code-analysis_part5/<h2 class="relative group">caffe中卷积运算的实现 <div id="caffe中卷积运算的实现" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#caffe%e4%b8%ad%e5%8d%b7%e7%a7%af%e8%bf%90%e7%ae%97%e7%9a%84%e5%ae%9e%e7%8e%b0" aria-label="Anchor">#</a> </span> </h2> <p>暴力实现的卷积大概是这样子的</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-07-caffe-source-code-analysis_part4/Tue, 07 Apr 2020 23:21:40 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-04-07-caffe-source-code-analysis_part4/<p>在看过 caffe 代码的三个核心部分 blob、layer、net 之后，陷入了不知道以什么顺序继续看的困境。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-01-12-caffe-source-code-analysis_part3/Sun, 12 Jan 2020 19:18:15 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-01-12-caffe-source-code-analysis_part3/<h2 class="relative group">Net 基本介绍 <div id="net-基本介绍" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#net-%e5%9f%ba%e6%9c%ac%e4%bb%8b%e7%bb%8d" aria-label="Anchor">#</a> </span> </h2> <p>网络通过组合与自动微分，共同定义一个函数及其梯度。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-01-11-caffe-source-code-analysis_part2/Sat, 11 Jan 2020 17:47:05 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-01-11-caffe-source-code-analysis_part2/<h2 class="relative group">layer 整体介绍 <div id="layer-整体介绍" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#layer-%e6%95%b4%e4%bd%93%e4%bb%8b%e7%bb%8d" aria-label="Anchor">#</a> </span> </h2> <p>layer 是模型计算的基本单元， 类似于 pytorch 或者其他深度学习框架的 op。 layer 中的数据流向为：输入若干个 blob，称之为 &ldquo;bottom blob&rdquo;，经过 layer 的计算后，输出若干个 blob，称之为 &ldquo;top blob&rdquo;。</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-01-10-caffe-source-code-analysis-part1/Fri, 10 Jan 2020 11:24:23 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/caffe-notes/2020-01-10-caffe-source-code-analysis-part1/<p>迫于生计，开始看 caffe 代码。 会侧重于分析 inference 部分。</p> <p>整体的依赖关系是：Net 由 Layer 组成，Layer 之间通过 Blob 传递数据。后面几篇会分别展开这三者。</p>