InterActive: Inter-Layer Activeness Propagation

摘要

An increasing number of computer vision tasks can be tackled with deepfeatures, which are the intermediate outputs of a pre-trained ConvolutionalNeural Network. Despite the astonishing performance, deep features extractedfrom low-level neurons are still below satisfaction, arguably because theycannot access the spatial context contained in the higher layers. In thispaper, we present InterActive, a novel algorithm which computes the activenessof neurons and network connections. Activeness is propagated through a neuralnetwork in a top-down manner, carrying high-level context and improving thedescriptive power of low-level and mid-level neurons. Visualization indicatesthat neuron activeness can be interpreted as spatial-weighted neuron responses.We achieve state-of-the-art classification performance on a wide range of imagedatasets.