1d_gradCAM

This repository provides a Class to perform grad-CAM [1] in Python with Keras library.

How to use

1. Object creation:
   The library is build following OOP paradigm. The first step is to create the object:

gCam = GradCam(testset, model)

where:

• testset is the testset on which compute gradCAM.
• model is the model to explain with gradCAM. It must be a keras model.

2. HeatMap computation:

heatmap = gCam.make_gradcam_heatmap(conv_layer_name)

where:

• conv_layer_name is the name of the convolutional layer on which applies the gradCAM. It is the same obtaine with the summary function of Keras.

3. Interpolation:
   It is possible to automatically interpolate the data to obtain the same dimension as the input feature. At the the moment there are three possible interpolation strategies:

• cubic spline interpolation, linear interpolation and matrix interpolation. For example the cubic spline interpolation is performed as follow:

  interpolate_vector = gCam.cubic_spline_interpolation(vector, v_size, input_size)

  where:
  • vector the vector to interpolate.
  • v_size size of the flatted heatmap. You have to multiply the two dimensions of the heatmap.
  • input_size size of the input