DK-CNNs: Dynamic kernel convolutional neural networks

This paper introduces dynamic kernel convolutional neural networks (DK-CNNs), an enhanced type of CNN, by performing line-by-line scanning regular convolution to generate a latent dimension weights. The proposed DK-CNN applies the DK weights, which rely on variable, and discretizes space variable extend new dimension; this process is named “DK convolution”. increases expressive capacity operation without increasing number parameters searching for useful patterns within extended dimension. In contrast conventional convolution, fixed analyse changed features, employs features. addition, can replace standard layer in any CNN network structure. DK-CNNs were compared with different structures CIFAR FashionMNIST datasets. experimental results show that achieve better performance than CNNs.