conv_visualiser

Python Package for Visualising Convolution Neural Network Features and Intermediate Activation of Tensorflow 2.0 Models

Reference

https://arxiv.org/abs/1610.02391
https://www.tensorflow.org/tutorials/interpretability/integrated_gradients

Requirements

Tensorflow 2.0 or newer
OpenCV
Matplotlib
numpy

Installation

pip install conv_visualiser

Usage

from cnn_visualiser import GRAD_CAM
from cnn_visualiser import GradVisualiser
from cnn_visualiser import IntermediateActivations
from cnn_visualiser import VanillaGradients
from cnn_visualiser import IntegratedGradients


from tensorflow.keras.applications import vgg16
from tensorflow.keras.applications.vgg16 import preprocess_input
from tensorflow.keras.preprocessing import image
import numpy as np
import matplotlib.pyplot as plt
import cv2

path = "./Input_Images/cat_dog.png" #path to image
vgg_model = vgg16.VGG16(weights = 'imagenet',input_shape=(224,224,3),include_top=True)
img = image.load_img(path,target_size=(224,224)) #make sure the dimensions of the image matches with the input_shape
img = image.img_to_array(img)
img = preprocess_input(img) #make sure the image is preprocessed
img = np.expand_dims(img,axis = 0)# adding a batch dimension


grad_cam = GRAD_CAM(model=vgg_model,input_image=img,layer_name='block5_conv3')
superimposed_image = grad_cam.get_superimposed(path) #heatmap superimposed on the original image
heatmap = grad_cam.get_heat_map()
guided_grads_cam = grad_cam.get_guided_grad_cam()

grad_visualiser = GradVisualiser(model=vgg_model,layer_name='block4_conv3',input_image=img)
guided_grads = grad_visualiser.guided_gradients()


vg = VanillaGradients(model = vgg_model,input_image = img)
filter_pattern = vg.generate_filter_pattern(filter_index=10,layer_name = 'block2_conv1',input_height=224,input_width=224)
grads = vg.vanilla_gradients(class_label = 245)


ia = IntermediateActivations(model=vgg_model,layer_names = ['block2_conv1','block2_conv2'])
ia.display_grid(input_image=img) #Matplotlib Plot

ia.display_single_channel(input_image = img)

ig = IntegratedGradients(model = vgg_model,input_image=img)
attributions = ig.integrated_gradients(batch_size=2) #to save  attributions
ig.display_plot_img_attributions(batch_size=2)# to display plot img