/
gradcamutils.py
executable file
·173 lines (138 loc) · 6.93 KB
/
gradcamutils.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
import numpy as np
import cv2
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras import backend as K
from tensorflow.keras.models import Model
import gc
tf.compat.v1.disable_eager_execution()
def normalize(x):
"""Utility function to normalize a tensor by its L2 norm"""
return (x + 1e-10) / (K.sqrt(K.mean(K.square(x))) + 1e-10)
def GradCam(model, img_array, layer_name):
cls = np.argmax(model.predict(img_array))
"""GradCAM method for visualizing input saliency."""
y_c = model.output[0, cls]
conv_output = model.get_layer(layer_name).output
grads = tf.gradients(y_c, conv_output)[0]
# grads = normalize(grads)
gradient_function = K.function([model.input], [conv_output, grads])
output, grads_val = gradient_function([img_array])
output, grads_val = output[0, :], grads_val[0, :, :, :]
weights = np.mean(grads_val, axis=(0, 1))
cam = np.dot(output, weights)
cam = np.maximum(cam, 0) # Passing through ReLU
cam /= np.max(cam) # scale 0 to 1.0
return cam
def GradCamPlusPlus(model, img_array, layer_name):
cls = np.argmax(model.predict(img_array))
y_c = model.output[0, cls]
conv_output = model.get_layer(layer_name).output
grads = tf.gradients(y_c, conv_output)[0]
# grads = normalize(grads)
first = K.exp(y_c)*grads
second = K.exp(y_c)*grads*grads
third = K.exp(y_c)*grads*grads*grads
gradient_function = K.function([model.input], [y_c,first,second,third, conv_output, grads])
y_c, conv_first_grad, conv_second_grad,conv_third_grad, conv_output, grads_val = gradient_function([img_array])
global_sum = np.sum(conv_output[0].reshape((-1,conv_first_grad[0].shape[2])), axis=0)
alpha_num = conv_second_grad[0]
alpha_denom = conv_second_grad[0]*2.0 + conv_third_grad[0]*global_sum.reshape((1,1,conv_first_grad[0].shape[2]))
alpha_denom = np.where(alpha_denom != 0.0, alpha_denom, np.ones(alpha_denom.shape))
alphas = alpha_num/alpha_denom
weights = np.maximum(conv_first_grad[0], 0.0)
alpha_normalization_constant = np.sum(np.sum(alphas, axis=0),axis=0)
alphas /= alpha_normalization_constant.reshape((1,1,conv_first_grad[0].shape[2]))
deep_linearization_weights = np.sum((weights*alphas).reshape((-1,conv_first_grad[0].shape[2])),axis=0)
cam = np.sum(deep_linearization_weights*conv_output[0], axis=2)
cam = np.maximum(cam, 0) # Passing through ReLU
cam /= np.max(cam) # scale 0 to 1.0
return cam
def softmax(x):
f = np.exp(x)/np.sum(np.exp(x), axis = 1, keepdims = True)
return f
def ScoreCam(model, img_array, layer_name, max_N=-1):
cls = np.argmax(model.predict(img_array))
act_map_array = Model(inputs=model.input, outputs=model.get_layer(layer_name).output).predict(img_array)
# extract effective maps
if max_N != -1:
act_map_std_list = [np.std(act_map_array[0,:,:,k]) for k in range(act_map_array.shape[3])]
unsorted_max_indices = np.argpartition(-np.array(act_map_std_list), max_N)[:max_N]
max_N_indices = unsorted_max_indices[np.argsort(-np.array(act_map_std_list)[unsorted_max_indices])]
act_map_array = act_map_array[:,:,:,max_N_indices]
input_shape = model.layers[0].output_shape[0][1:] # get input shape
# 1. upsampled to original input size
act_map_resized_list = [cv2.resize(act_map_array[0,:,:,k], input_shape[:2], interpolation=cv2.INTER_LINEAR) for k in range(act_map_array.shape[3])]
# 2. normalize the raw activation value in each activation map into [0, 1]
act_map_normalized_list = []
for act_map_resized in act_map_resized_list:
if np.max(act_map_resized) - np.min(act_map_resized) != 0:
act_map_normalized = act_map_resized / (np.max(act_map_resized) - np.min(act_map_resized))
else:
act_map_normalized = act_map_resized
act_map_normalized_list.append(act_map_normalized)
# 3. project highlighted area in the activation map to original input space by multiplying the normalized activation map
masked_input_list = []
for act_map_normalized in act_map_normalized_list:
masked_input = np.copy(img_array)
for k in range(3):
masked_input[0,:,:,k] *= act_map_normalized
masked_input_list.append(masked_input)
masked_input_array = np.concatenate(masked_input_list, axis=0)
# 4. feed masked inputs into CNN model and softmax
pred_from_masked_input_array = softmax(model.predict(masked_input_array))
# 5. define weight as the score of target class
weights = pred_from_masked_input_array[:,cls]
# 6. get final class discriminative localization map as linear weighted combination of all activation maps
cam = np.dot(act_map_array[0,:,:,:], weights)
cam = np.maximum(0, cam) # Passing through ReLU
cam /= np.max(cam) # scale 0 to 1.0
return cam
def superimpose(original_img_path, cam, emphasize=False):
img_bgr = cv2.imread(original_img_path)
heatmap = cv2.resize(cam, (img_bgr.shape[1], img_bgr.shape[0]))
if emphasize:
heatmap = sigmoid(heatmap, 50, 0.5, 1)
heatmap = np.uint8(255 * heatmap)
heatmap = cv2.applyColorMap(heatmap, cv2.COLORMAP_JET)
hif = .8
superimposed_img = heatmap * hif + img_bgr
superimposed_img = np.minimum(superimposed_img, 255.0).astype(np.uint8) # scale 0 to 255
superimposed_img_rgb = cv2.cvtColor(superimposed_img, cv2.COLOR_BGR2RGB)
return superimposed_img_rgb
import tensorflow.keras
import tensorflow as tf
from tensorflow.python.framework import ops
from tensorflow.keras.applications.vgg16 import preprocess_input
def build_guided_model(build_model_function):
"""Function returning modified model.
Changes gradient function for all ReLu activations according to Guided Backpropagation.
"""
if "GuidedBackProp" not in ops._gradient_registry._registry:
@ops.RegisterGradient("GuidedBackProp")
def _GuidedBackProp(op, grad):
dtype = op.inputs[0].dtype
return grad * tf.cast(grad > 0., dtype) * \
tf.cast(op.inputs[0] > 0., dtype)
g = tf.compat.v1.get_default_graph()
with g.gradient_override_map({'Relu': 'GuidedBackProp'}):
new_model = build_model_function()
return new_model
def GuidedBackPropagation(model, img_array, layer_name):
model_input = model.input
layer_output = model.get_layer(layer_name).output
max_output = K.max(layer_output, axis=3)
grads = tf.gradients(max_output, model_input)[0]
get_output = K.function([model_input], [grads])
saliency = get_output([img_array])
saliency = np.clip(saliency[0][0], 0.0, 1.0) # scale 0 to 1.0
return saliency
def sigmoid(x, a, b, c):
return c / (1 + np.exp(-a * (x-b)))
from tensorflow.keras.preprocessing.image import load_img, img_to_array
def read_and_preprocess_img(path, size=(224,224)):
img = load_img(path, target_size=size)
x = img_to_array(img)
x = np.expand_dims(x, axis=0)
x = preprocess_input(x)
return x