Selective Kernel Networks

CVPR 2019 Xiang LiWenhai WangXiaolin HuJian Yang

In standard Convolutional Neural Networks (CNNs), the receptive fields of artificial neurons in each layer are designed to share the same size. It is well-known in the neuroscience community that the receptive field size of visual cortical neurons are modulated by the stimulus, which has been rarely considered in constructing CNNs... (read more)

PDF Abstract

Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Classification CIFAR-10 SKNet-29 (ResNeXt-29, 16×32d) Percentage correct 96.53 # 29
Image Classification CIFAR-100 SKNet-29 (ResNeXt-29, 16×32d) Percentage correct 82.67 # 20
Image Classification ImageNet SKNet-101 Top 1 Accuracy 79.81% # 65
Number of params 48.9M # 26

Methods used in the Paper