Learning Deep Features for Discriminative Localization [article]

Bolei Zhou and Aditya Khosla and Agata Lapedriza and Aude Oliva and Antonio Torralba
2015 arXiv   pre-print
In this work, we revisit the global average pooling layer proposed in [13], and shed light on how it explicitly enables the convolutional neural network to have remarkable localization ability despite being trained on image-level labels. While this technique was previously proposed as a means for regularizing training, we find that it actually builds a generic localizable deep representation that can be applied to a variety of tasks. Despite the apparent simplicity of global average pooling, we
more » ... are able to achieve 37.1% top-5 error for object localization on ILSVRC 2014, which is remarkably close to the 34.2% top-5 error achieved by a fully supervised CNN approach. We demonstrate that our network is able to localize the discriminative image regions on a variety of tasks despite not being trained for them
arXiv:1512.04150v1 fatcat:zkorupnj6jhi7n2c7lswueg5gq