可视化_CNN


Grad-CAM: VisualExplanationsfromDeepNetworksviaGradient-basedLocalization
Deconvolution:Visualizing and Understanding Convolutional Networks
       从本质上说,反卷积和导向反向传播的基础都是反向传播,二者唯一的区别在于反向传播过程中经过ReLU层时对梯度的不同处理策略。
1、CNN      2、可解释性      3、反卷积(Deconvolution)      4、导向反向传播(Guided-backpropagation)
Github:Tensorflow_Grad_CAM_plus_plus
              Grad-CAM-tensorflow
Paper:Grad-CAM++
           Guided-Backpropagation
           Visualizing and Understanding Convolutional Networks

|