Exploiting Features with Split-and-Share Module

Deep convolutional neural networks (CNNs) have shown state-of-the-art performances in various computer vision tasks. Advances on CNN architectures focused mainly designing blocks of the feature extractors, but less classifiers that exploit extracted features. In this work, we propose Split-and-Share Module (SSM), a classifier splits given into parts, which are partially shared by multiple sub-classifiers. Our intuition is more features shared, common they will become, and SSM can encourage such structural characteristics split be easily integrated any architecture without bells whistles. We extensively validated efficacy ImageNet-1K classification task, has consistent significant improvements over baseline architectures. addition, analyze effect using Grad-CAM visualization.