SA-Net
SA-Net copied to clipboard
Code for our ICASSP 2021 paper: SA-Net: Shuffle Attention for Deep Convolutional Neural Networks
请问一下论文里面提到的. First, we use Group Norm (GN)[25 ] over Xk2 to obtain spatial-wise statistics.,为什么通过GN就可以获取到空间注意力啊
本文中的SAnet是否可以去替换有的模型中使用的SCnet,就是说接在图像Bev特征后面去对齐lidar的bev特征呢?
你好,请问sa模块是否可以作为即插即用的子模块,用在特征图之后的其他任务分支上?如使用resnet提取特征之后,特征图在不同head上的操作中加入sa模块
如题所示,我看说G通常的32或64,这个是怎么来的
热力图问题
请问一下,作者能公开一下,热力图的实现代码吗?十分感谢
hello,can u give the password of pre-trained weight from baiduwangpan? I will Thank u very much
when using sa_layer in deeplearning model getting following error, kindly help to me, solve this x = x.reshape(b * self.groups, -1, h, w) RuntimeError: shape '[64, -1, 64, 64]' is...
我在yolov5中插入sanet,出现错误 x_0, x_1 = x.chunk(2, dim=1) ValueError: not enough values to unpack (expected 2, got 1) 因此我打印出他们的形状 def forward(self, x): b, c, h, w = x.shape print(x.shape) x = x.reshape(b...
我看代码中默认为64,那么是否50/4就是差不多的groups数量