<article style="font-size: 16px;">
<div>
<section>
<div>
<div>
<p>This is my first blog and I have decided to write a small description about one of the activation functions used in the <strong>YOLOv4</strong>. <strong>YOLO</strong>, or <strong>You Only Look Once</strong>, is a one-shot object detection technique that was introduced by Joseph Redmon and Ali Farhadi in 2016 and there are already 4 versions of the technique. Here, we will take a look at YOLOv4, specifically its performance optimizers, the two <strong>“Bags” </strong>of optimization functions used: the <strong>“Bag of Freebies (BoF)” </strong>which is used during the training time and <strong>“Bag of Specials (BoS)”</strong>,<strong> </strong>used during the inference.</p>
<p> 这是我的第一个博客，我决定对<strong>YOLOv4中</strong>使用的一种激活功能进行<strong>简短描述</strong> 。 <strong>YOLO</strong>或“ <strong>您只看一次</strong> ”是一种单发物体检测技术，该技术由约瑟夫·雷德蒙(Joseph Redmon)和阿里·法哈迪(Ali Farhadi)于2016年推出，目前已经有4种版本。 在这里，我们将介绍YOLOv4，特别是其性能优化器，其中使用了两个优化功能的<strong>“包”</strong> ：培训期间使用的<strong>“赠品包(BoF)”</strong>和<strong>“特价包(BoS)”</strong> ， <strong> </strong> 在推断过程中使用。 </p>
<p>The Bag of Specials contains low computational cost modules for both the backbone and the detector of the YOLOv4 architecture. These are:</p>
<p> 特价袋包含用于YOLOv4架构的主干和检测器的低计算成本模块。 这些是： </p>
<p>Here, we can see that the Mish activation function is present in both the backbone and the detector. So, what makes it “special”? Let’s understand a bit more about this activation function.</p>
<p> 在这里，我们可以看到Mish激活功能同时存在于主干和检测器中。 那么，是什么使它成为“特殊”的呢？ 让我们进一步了解此激活功能。 </p>
</div>
</div>
</section>
<section>
<div>
<div>
<h1> <strong>碗碟激活功能：</strong> <span style="font-weight: bold;">(</span><strong>Mish Activation Function:</strong><span style="font-weight: bold;">)</span></h1>
<p><strong>Mish </strong>is a smooth, non-monotonic activation function, that can be defined as:</p>
<p> <strong>Mish</strong>是一种平滑的，非单调的激活函数，可以定义为： </p>
<p><strong>f(x) = x</strong><strong>tanh((x))</strong></p>
<p> <strong>f(x)= x</strong><strong>tanh((x))</strong> </p>
<p>where, <strong>(x) = ln(1+e^x)</strong>, is a <strong>softmax</strong> activation function.</p>
<p> 其中<strong>(x)= ln(1 + e ^ x)</strong>是<strong>softmax</strong>激活函数。 </p>
<figure style="display:block;text-align:center;">
<div>
<div>
<div>
<div>
<div style="text-align: center;">
<img alt="Image for post" height="456" src="https://beijingoptbbs.oss-cn-beijing.aliyuncs.com/cs/5606289-d83f33ef8dc26f1b43fe1222440c793e.png" style="outline: none;" width="816">
</div>
</div>
</div>
</div>
</div>
</figure>
<figure style="display:block;text-align:center;">
<div>
<div>
<div>
<div style="text-align: center;">
<img alt="Image for post" height="29" src="https://beijingoptbbs.oss-cn-beijing.aliyuncs.com/cs/5606289-e7533ad7165ef0089dbd426f8423aa89.png" style="outline: none;" width="181">
</div>
</div>
</div>
</div>
<figcaption>
Source: YOLOv4 — Part 3: Bag of Specials | VisionWizard
</figcaption>
<figcaption>
来源：YOLOv4 —第3部分：特价商品| 视觉向导
</figcaption>
</figure>
<p>This is very similar to another activation function called <strong>Swish</strong> function, that can be defined as:</p>
<p> 这与另一个称为<strong>Swish</strong>函数的激活函数非常相似，可以将其定义为： </p>
<figure style="display:block;text-align:center;">
<div>
<div>
<div>
<div style="text-align: center;">
<img alt="Image for post" height="288" src="https://beijingoptbbs.oss-cn-beijing.aliyuncs.com/cs/5606289-7a6484dfd4f1bd13bafd19ba82a9f808.png" style="outline: none;" width="550">
</div>
</div>
</div>
</div>
</figure>
<figure style="display:block;text-align:center;">
<div>
<div>
<div>
<div style="text-align: center;">
<img alt="Image for post" height="30" src="https://beijingoptbbs.oss-cn-beijing.aliyuncs.com/cs/5606289-c8de0bbf8c26de308f5cfa428339c9bd.png" style="outline: none;" width="194">
</div>
</div>
</div>
</div>
<figcaption>
Source: Swish: Booting ReLU from the Activation Function Throne
</figcaption>
<figcaption>
来源：Swish：从激活功能王座引导ReLU
</figcaption>
</figure>
<p>The reason why Mish function is used in YOLOv4 is because of its low cost and its various properties like it’s smooth and non-monotonic nature, unbo |
|
转播
