How to use SSDlayer in MATLAB to create SSD

最新推荐文章于 2024-04-08 12:30:02 发布

原创最新推荐文章于 2024-04-08 12:30:02 发布 · 244 阅读

0 ·

本内容遵循CC 4.0 BY-SA版权协议

标签

#matlab #开发语言

收录于

deeplearning matlab 专栏收录该内容

41 篇文章

订阅专栏

博主在尝试使用SSD（Single Shot Multibox Detector）重新训练模型时遇到问题，发现效果不佳。问题在于没有添加anchor boxes，这是SSD架构的关键组成部分。通过查阅帮助文档，博主了解到构建SSD的具体步骤，包括设置特征提取网络、输入尺寸、类别数、基础网络、anchor boxes和连接的层。尽管如此，文档的指导仍然不够充分。博主分享了自己尝试创建SSD层的代码，并邀请读者查看详细的SSD层实现细节，寻求进一步的帮助和讨论。

I have tried to use SSD to retrain my model, but it doesn't work well.

and realize that I have not adding anchor boxes.

However, the help document is not enough to build the ssd.

You can follow what I did in the following to create the layer.

%% SSD with anchor
featureExtractionNetwork = resnet50;% very inportant here

%lgraph = ssdLayers(inputSize,1,featureExtractionNetwork)
%lgraph = ssdLayers(imageSize, numClasses, baseNetwork, anchorBoxes, layersToConnect);
anchorBoxes = {[ 52    14;    52    11;    41    13;    46    10;    40     7;     3     1]};
featureLayer = "activation_43_relu";%important here
lgraph= ssdLayers(inputSize, 1, featureExtractionNetwork, anchorBoxes,featureLayer)
analyzeNetwork(lgraph)