划水

每日Attention学习27——Patch-based Graph Reasoning[NC 25] [link] Graph-based context learning network for infrared small target detection

每日Attention学习24——Strip Convolution Block[TIP 21] [link] CoANet: Connectivity Attention Network for Road Extraction From Satellite Imagery

每日Attention学习23——KAN-Block[SPL 25] [link] [code] KAN See In the DarkKolmogorov-Arnold Network Block (KAN-Block)

每日Attention学习19——Convolutional Multi-Focal Attention每日Attention学习19——Convolutional Multi-Focal Attention

每日Attention学习18——Grouped Attention Gate[ICLR 25 Submission] [link] UltraLightUNet: Rethinking U-shaped Network with Multi-kernel Lightweight Convolutions for Medical Image Segmentation

Pytorch基于VGG cosine similarity实现简单的以图搜图(图像检索)代码如下：上述代码的核心思想类似于感知损失(Perceptual Loss)，利用VGG提取图像的多级特征，从而比较两张图像之间的相似性。区别在于Perceptual Loss中一般使用MAE，MSE比较特征的距离，而这里的代码使用余弦相似度。