MatConvNet Saliency Visualization
This is a MatConvNet demo of several saliency visualization methods of ConvNet models.
References
- Error backpropagation: Simonyan, K., Vedaldi, A., Zisserman, A.: Deep inside convolutional networks: Visualising image classification models and saliency maps.
- Class Activation Map: Zhou, B., Khosla, A., Lapedriza, A., Oliva, A [[GitHub](https://github.com/jimmie33/Caffe-ExcitationBP]., Torralba, A.: Learning deep features for discriminative localization. [GitHub]
- Excitation backpropagation: Jianming Zhang, Zhe Lin, Jonathan Brandt, Xiaohui Shen, Stan Sclaroff. Top-down Neural Attention by Excitation Backprop.) [GitHub]
Prerequisites
- MatConvNet.
- A trained ConvNet model e.g. ResNet-152 or others here.
Installation
- Compile MatConvNet.
- Download the model from the links above, the default model of the demo is the ResNet-152 and place in
/Models
. - Replace the default Conv.m and Pooling.m files in the default MatConvNet folder
/MatConvNet/matlab/+dagnn
with the ones included.