Inspired by the recent neuroscience studies on the left-right asymmetry of the brains in the low and high spatial frequency processing, we introduce a novel type of network -- the left-right skip-densenets for coarse-to-fine object categorization. This network can enable both coarse and fine-grained classification in a single framework. We also for the first time propose the layer-skipping mechanism which learns a gating network to predict whether skip some layers in the testing stage. This layer-skipping mechanism assigns more flexibility and capability to our network for the categorization tasks. Our network is evaluated on three widely used datasets; the results show that our network is more promising in solving the coarse-to-fine object categorization than the competitors.

本文介绍了一种双重跳过网络，该网络由于最近神经科学研究在处理低和高空间频率信息的人类大脑的左右不对称性，这对于粗到细的对象分类具有两个分支。为了增加模型的灵活性和实用性，本文提出了一种学习绕过层的门控网络的层绕过机制。在多个常用的粗到细级别目标分类基准上进行评估，证明了所提出的网络模型具有良好的性能和灵活性。

双跳网络