Learning Deep Features for Discriminative Localization

Code

• github.com/zhoubolei/CAM
  Officialtf★ 0
• github.com/FrancescoSaverioZuppichini/A-journey-into-Convolutional-Neural-Network-visualization-
  pytorch★ 275
• github.com/joaopauloschuler/k-neural-api
  tf★ 135
• github.com/innat/ML-Resource
  tf★ 113
• github.com/Azure/AzureChestXRay
  pytorch★ 88
• github.com/zdcuob/Fully-Convlutional-Neural-Networks-for-state-of-the-art-time-series-classification-
  tf★ 12
• github.com/vlue-c/PyTorch-Explanations
  pytorch★ 7
• github.com/windstormer/Cfd-CAM
  pytorch★ 3
• github.com/Seb-Good/deep_ecg
  tf★ 0
• github.com/sauravmishra1710/EXPLAINABLE-AI---Skin-Cancer-Detection-explained-with-GRADCAM
  tf★ 0

Abstract

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 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

Tasks

Reproductions