FeatureMapVisualization

The same class, FeatureMapVisualization, is available in both python and matlab versions. This class provides a way to visualize the feature maps of deep learning networks.

More details (introduction of parameters, etc.) are described in the class documentation.

---

Python

This package provides a class that can be used to visualize feature maps at various layers of a deep learning network.

How to use the FeatureMapVisualization class:

Initialization flattens the network structure and stores the layers sequentially in a list.

import FeatureMapVisualization as Fmv

model = models.alexnet(pretrained=True)

modelLayer = list(model.children())

modelVisualization = Fmv.FeatureMapVisualization(modelLayer)

---

Provides functions and methods:

This method displays the layers of the flattened network:

modelVisualization.showStructure()

This method shows the visualization of the feature map of the specified layer:

modelVisualization.visualization(input,k=0,type='pdf',resolution=(256,256))

This method shows the visualization of all layer feature maps:

modelVisualization.visualizationEveryLayer(input,type='pdf',resolution=(256,256))

---

Matlab

How to use the FeatureMapVisualization class:

Feed in a model for instantiation and initialization.

a = alexnet;
Fmv = FeatureMapVisualization(a,isShow=true);

---

Provides functions and methods:

This method shows the visualization of the feature map of the specified layer:

Fmv.visualization(input,k=2,cmap=hsv,resolution=[256,256]);

This method shows the visualization of all layer feature maps:

Fmv.visualizationEveryLayer(input,cmap=gray,isSave=false);