<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[qq_36070656的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/qq_36070656</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; qq_36070656]]></copyright><item><title><![CDATA[【YOLO改进】Efficient Decoupled Head解耦合头的添加]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/134074092</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/134074092</guid><author>qq_36070656</author><pubDate>Fri, 27 Oct 2023 13:39:38 +0800</pubDate><description><![CDATA[添加解耦合头]]></description><category></category></item><item><title><![CDATA[YOLO改进：【损失函数篇】提高小目标检测精度——Normalized Wasserstein Distance (NWD)损失函数]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/133910972</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/133910972</guid><author>qq_36070656</author><pubDate>Wed, 18 Oct 2023 19:40:54 +0800</pubDate><description><![CDATA[NWD损失函数助力小目标检测]]></description><category></category></item><item><title><![CDATA[YOLO改进：动态卷积之有效推理的条件参数化卷积CondConv]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/133889191</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/133889191</guid><author>qq_36070656</author><pubDate>Wed, 18 Oct 2023 15:21:31 +0800</pubDate><description><![CDATA[动态卷积：CondConv在YOLO中的应用]]></description><category></category></item><item><title><![CDATA[YOLO改进：添加额外的检测头]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/133824066</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/133824066</guid><author>qq_36070656</author><pubDate>Sat, 14 Oct 2023 13:21:45 +0800</pubDate><description><![CDATA[在YOLO中添加浅层检测头，提高对小目标的检测能力，并在VisDrone数据集中进行测试]]></description><category></category></item><item><title><![CDATA[YOLO改进：SPD模块--不再跨行卷积或池化:低分辨率图像和小对象的新CNN构建块]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/133771824</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/133771824</guid><author>qq_36070656</author><pubDate>Wed, 11 Oct 2023 16:06:20 +0800</pubDate><description><![CDATA[SPD-Conv可以防止小目标细粒度信息的丢失]]></description><category></category></item><item><title><![CDATA[YOLO改进：轻量级通用上采样算子CARAFE]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/133760323</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/133760323</guid><author>qq_36070656</author><pubDate>Wed, 11 Oct 2023 09:49:14 +0800</pubDate><description><![CDATA[使用CARAFE算子改进YOLOv5模型，将YOLOv5中的所有上采样操作替换，在保持较少参数增加的同时，提高模型的检测精度。在VisDrone数据集中进行测试。]]></description><category></category></item><item><title><![CDATA[YOLO改进：一种级联与并联结合的多尺度特征融合模块RFEM]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/133753811</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/133753811</guid><author>qq_36070656</author><pubDate>Tue, 10 Oct 2023 19:13:42 +0800</pubDate><description><![CDATA[该模块是由中提出。具体来说，输入特征图将经过三个不同空洞率（2，4，8）的空洞卷积，并依次进行通道拼接。该模块的优势在于，能够有效地捕获平衡的非局部上下文特征和局部目标特征，解决小目标局部语义缺失。如果F的分辨率是62*62，且感受野为1，那么R1、R2、R3、P的感受野分别是5、13、29、29。P的感受野达到F的一半大小，这将使得输出特征图包含足够的非局部背景信息。中的RFEM，其使用三个并联的空洞卷积融合不同尺度的特征信息。]]></description><category></category></item><item><title><![CDATA[YOLO改进：添加多尺度特征提取模块Scale-Aware RFE Model]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/133670794</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/133670794</guid><author>qq_36070656</author><pubDate>Sun, 08 Oct 2023 11:16:54 +0800</pubDate><description><![CDATA[RFEM的结构很简单，该结构使用了三个不同空洞率（1、2、3）的空洞卷积提取特征以提取多尺度信息；此外，不同分支共享权重减少参数量；另外，使用残差连接防止梯度爆炸的问题；最后，将四个分支的特征相加得到输出的特征层。]]></description><category></category></item><item><title><![CDATA[ORB-SLAM3使用压缩图（ROS）]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/132766710</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/132766710</guid><author>qq_36070656</author><pubDate>Fri, 08 Sep 2023 20:04:35 +0800</pubDate><description><![CDATA[ORB-SLAM3中使用的是，这是一种无损的图片格式，通常使用D435等相机采集图片的时候，会导致rosbag包巨大。因此，在实际中，会采集压缩格式的。具体以ros中的中为例子。]]></description><category></category></item><item><title><![CDATA[Yolov5中使用EigenCAM实现特征图可视乎]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/131740372</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/131740372</guid><author>qq_36070656</author><pubDate>Sat, 15 Jul 2023 16:15:35 +0800</pubDate><description><![CDATA[是的，我了解EigenCAM，它是一种用于使卷积神经网络(CNN)模型更具解释性的方法。EigenCAM的主要思想是：通过层析分解的方法找出输入图像在CNN中间层特征图中激活程度最高的区域。将中间层特征图中每个特征通道的激活作为输入图像不同区域的重要性权重。权重映射回输入图像，就可以高亮显示对CNN分类决策影响最大的图像区域。相比普通的类别激活映射(CAM)，EigenCAM具有以下优点:可以用于任意的CNN模型，无需修改模型结构。计算效率高，可以快速生成类激活映射。]]></description><category></category></item><item><title><![CDATA[DCGAN的工程化]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/131689217</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/131689217</guid><author>qq_36070656</author><pubDate>Wed, 12 Jul 2023 20:00:23 +0800</pubDate><description><![CDATA[为DCGAN添加了日志、每x个epoch保存模型、绘制loss曲线、生成对比图功能，方便使用。]]></description><category></category></item><item><title><![CDATA[在Jetson Xavier NX配置Yolov5s环境]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/129679994</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/129679994</guid><author>qq_36070656</author><pubDate>Tue, 21 Mar 2023 15:30:28 +0800</pubDate><description><![CDATA[df -htop。]]></description><category></category></item><item><title><![CDATA[【SLAM】从零开始在Ubuntu系统中复现ORB-SLAM3（version1.0）]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/129499785</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/129499785</guid><author>qq_36070656</author><pubDate>Mon, 13 Mar 2023 22:38:46 +0800</pubDate><description><![CDATA[ubuntu20中基于ROS的orb-slam3]]></description><category></category></item><item><title><![CDATA[【AirSim】从零开始在Ubuntu系统中搭建Unreal4和AirSim环境]]></title><link>https://blog.youkuaiyun.com/qq_36070656/article/details/129490511</link><guid>https://blog.youkuaiyun.com/qq_36070656/article/details/129490511</guid><author>qq_36070656</author><pubDate>Mon, 13 Mar 2023 13:28:44 +0800</pubDate><description><![CDATA[基于Ubuntu20下使用AirSim仿真]]></description><category></category></item></channel></rss>