TL;DR本文提出了一种基于本地重要性建模的简单,通用且有效的池化层 LIP,它能够学习自适应重要性权重,从而提高下采样过程中的判别特征,在 ImageNet 分类和 MS COCO 数据集的目标检测中都取得了很好的性能。
Abstract
Spatial downsampling layers are favored in convolutional neural networks
(CNNs) to downscale feature maps for larger receptive fields and less memory
consumption. However, for discriminative tasks, there is a possibility that
these layers lose the discriminative details due to improper