Hay-Blog

https://nodequality.com/r/AsCpkI9MBxAklxGexeButfkC68omVUy7

https://nodequality.com/r/91XXNNKy6jrn9mn6yp27mNWEE4M6VMrY

【ICML-2021】SimAM架构能量函数 最小化能量函数的方程 最终的输出结果 \begin{aligned} \frac{1}{e} = \frac{(t-\hat \mu)^2}{4(\hat \sigma^2 + \lambda)} + \frac{2(\hat \sigma^2 + \lambda)}{4(\hat \sigma^2 + \lambda)} = \frac{(t-\hat \mu)^2}{4(\hat \sigma^2 + \lambda)} + \frac {1}{2} \end{aligned}实现1234567891011121314151617181920212223242526import torchfrom torch import nnimport torch.nn.functional as Fclass SimAM(nn.Module): def __init__(self,Lambda=0.1): """ :param Lambda: 能量函数里面的超参数 ...

【CVPR-21】Coordinate Attention架构 实现123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051import torchfrom torch import nnimport torch.nn.functional as Fclass CoordinateAttention(nn.Module): def __init__(self,in_ch,r): """ :param in_ch: 特征图输入通道数 :param r: 特征图通道数的减少比率 """ super(CoordinateAttention, self).__init__() self.x_avg_pool = nn.AdaptiveAvgPool2d((None,1)) self.y_avg_pool ...

【CVPR-20】ECANet架构输入F，[N,C,H,W] 经过GAP，[N,C,H,W]→[N,C,1,1]→[N,C,1]→[N,1,C] 经过卷积，[N,1,C]→[N,1,C]→[N,C,1]→[N,C,1,1] 经过逐元素相乘，[N,C,1,1]→[N,C,H,W] 实现 123456789101112131415161718192021222324252627282930import torchfrom torch import nnclass ECANet(nn.Module): def __init__(self,in_ch,gamma=2,b=1): super(ECANet, self).__init__() self.gap = nn.AdaptiveAvgPool2d(1) self.in_ch = torch.tensor(in_ch,dtype=torch.float32) self.k = int(abs(torch.log2(self.in_ch) + b) / gamma) ...

【ECCV-18】CBAM架构输入特征图F（N,C,H,W） 经过CA模块转化为F’（N,C,H,W）* CA（N,C,1,1）→ （N,C,H,W） 经过SA模块转化为F’’（N,C,H,W）*SA（N,1,H,W）→（N,C,H,W） 最后残差连接 实现CA模块 1234567891011121314151617181920212223242526272829303132333435class ChannelAttentionModule(nn.Module): def __init__(self,in_ch,reduction_ratio): """ :param in_ch: 特征图输入通道数 :param reduction_ratio: 特征图通道数的减少比率 """ super(ChannelAttentionModule, self).__init__() self.max_pool = nn.AdaptiveMax ...

[CVPR-18]SENets架构 实现12345678910111213141516171819202122232425262728293031323334353637383940414243import torchfrom torch import nnclass SENets(nn.Module): def __init__(self,in_ch,reduction_ratio): super(SENets, self).__init__() """ :param in_ch: 特征图输入通道数 :param reduction_ratio: 特征图通道数的减少比率 """ # [N,C,H,W] -> [N,C,1,1] self.Global_pooling = nn.AdaptiveAvgPool2d(1) self.fc = nn.Sequential( ...

Dataset-CelebA主要包括transforms数据增强、DatasetFolder加载无分类数据 transforms数据增强123456transforms = transforms.Compose([ transforms.Resize(64), transforms.CenterCrop(64), transforms.ToTensor(), transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))]) DatasetFolder加载无分类数据1234567# 数据加载（无分类数据）celeba_set = DatasetFolder( root=win_root, loader=loader, extensions="jpg", transform= transforms) 完整代码12345678910111213141516171819202122232425from torchvision.datasets import Da ...

Docker部署后端及配置集群1、后端项目打包1mvn clean install -Dmaven.test.skip=true 2、安装java镜像1docker pull openjdk 3、创建并启动java容器1234567891011#创建数据卷，上传JAR文件docker volume create j1#启动容器docker run -it -d --name j1 -v j1:/home/soft --net=host openjdk#进入j1容器docker exec -it j1 bash#启动Java项目nohup java -jar /home/soft/你的jar包名字.jar >> home/soft/java.log 2>&1 &#查看日志tail -100f /home/soft/java.log .mjtocborvham{zoom:50%;} 12345678910111213#同理部署两个后端节点，记得修改好配置文件端口号docker volume create j2docker run -it -d --name ...

Docker配置RocketMQ1、拉取RocketMQ镜像 1docker pull foxiswho/rocketmq:broker-4.5.1 2、启动注册中心Nameserver 1docker run -d -p 9876:9876 --name rmqserver foxiswho/rocketmq:server-4.5.1 3、创建RocketMQ配置文件 .rxhfvogxhskb{zoom: 50%;} 12//或者本地创建后上传到云主机上，路径可以根据自己设定vim /usr/local/roketmq/conf/broker.conf 123456789brokerIP1 = 填写你的云主机ipbrokerClusterName = DefaultClusterbrokerName = broker-abrokerId = 0deleteWhen = 04fileReservedTime = 48brokerRole = ASYNC_MASTERflushDiskType = ASYNC_FLUSHenablePropertyFilter=true 4、启动Roc ...