-
Notifications
You must be signed in to change notification settings - Fork 0
/
Grad-CAM++(may be errors).py
185 lines (163 loc) · 6.48 KB
/
Grad-CAM++(may be errors).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
174
175
176
177
178
179
180
181
182
183
184
185
def target_category_loss(x, category_index, nb_classes):
return torch.mul(x, F.one_hot(category_index, nb_classes))
def target_category_loss_output_shape(input_shape):
return input_shape
def normalize(x):
# utility function to normalize a tensor by its L2 norm
return x / (torch.sqrt(torch.mean(torch.square(x))) + 1e-5)
def resize_1d(array, shape):
res = np.zeros(shape)
if array.shape[0] >= shape:
ratio = array.shape[0]/shape
for i in range(array.shape[0]):
res[int(i/ratio)] += array[i]*(1-(i/ratio-int(i/ratio)))
if int(i/ratio) != shape-1:
res[int(i/ratio)+1] += array[i]*(i/ratio-int(i/ratio))
else:
res[int(i/ratio)] += array[i]*(i/ratio-int(i/ratio))
res = res[::-1]
array = array[::-1]
for i in range(array.shape[0]):
res[int(i/ratio)] += array[i]*(1-(i/ratio-int(i/ratio)))
if int(i/ratio) != shape-1:
res[int(i/ratio)+1] += array[i]*(i/ratio-int(i/ratio))
else:
res[int(i/ratio)] += array[i]*(i/ratio-int(i/ratio))
res = res[::-1]/(2*ratio)
array = array[::-1]
else:
ratio = shape/array.shape[0]
left = 0
right = 1
for i in range(shape):
if left < int(i/ratio):
left += 1
right += 1
if right > array.shape[0]-1:
res[i] += array[left]
else:
res[i] += array[right] * \
(i - left * ratio)/ratio+array[left]*(right*ratio-i)/ratio
res = res[::-1]
array = array[::-1]
left = 0
right = 1
for i in range(shape):
if left < int(i/ratio):
left += 1
right += 1
if right > array.shape[0]-1:
res[i] += array[left]
else:
res[i] += array[right] * \
(i - left * ratio)/ratio+array[left]*(right*ratio-i)/ratio
res = res[::-1]/2
array = array[::-1]
return res
class ActivationsAndGradients:
""" Class for extracting activations and
registering gradients from targetted intermediate layers """
def __init__(self, model, target_layer):
self.model = model
self.gradients = []
self.activations = []
target_layer.register_forward_hook(self.save_activation)
target_layer.register_backward_hook(self.save_gradient)
def save_activation(self, module, input, output):
self.activations.append(output)
def save_gradient(self, module, grad_input, grad_output):
# Gradients are computed in reverse order
self.gradients = [grad_output[0]] + self.gradients
def __call__(self, x):
self.gradients = []
self.activations = []
return self.model(x)
class BaseCAM:
def __init__(self, model, target_layer, use_cuda=False):
self.model = model.eval()
self.target_layer = target_layer
self.cuda = use_cuda
if self.cuda:
self.model = model.cuda()
self.activations_and_grads = ActivationsAndGradients(self.model, target_layer)
def forward(self, input_img):
return self.model(input_img)
def get_cam_weights(self,
input_tensor,
target_category,
activations,
grads):
raise Exception("Not Implemented")
def get_loss(self, output, target_category):
print(output.size())
return output[target_category]
def __call__(self, input_tensor, target_category=None):
if self.cuda:
input_tensor = input_tensor.cuda()
output = self.activations_and_grads(input_tensor)
if target_category is None:
output = output.squeeze()
target_category = np.argmax(output.cpu().data.numpy())
print(output)
print(target_category)
self.model.zero_grad()
loss = self.get_loss(output, target_category)
loss.backward(retain_graph=True)
activations = self.activations_and_grads.activations[-1].cpu().data.numpy()[0, :]
grads = self.activations_and_grads.gradients[-1].cpu().data.numpy()[0, :]
#weights = np.mean(grads, axis=(0))
weights = self.get_cam_weights(input_tensor, target_category, activations, grads)
cam = np.zeros(activations.shape[1:], dtype=np.float32)
#
for i, w in enumerate(weights):
cam += w * activations[i, :]
# cam = activations.dot(weights)
# cam = activations.dot(weights)
# print(input_tensor.shape[1])
# print(cam.shape)
# x = np.arange(0, 247, 1)
# plt.plot(x, cam.reshape(-1, 1))
# sns.set()
# ax = sns.heatmap(cam.reshape(-1, 1).T)
#cam = cv2.resize(cam, input_tensor.shape[1:][::-1])
cam = resize_1d(cam, (input_tensor.shape[2]))
#cam = np.maximum(cam, 0)
# cam = np.expand_dims(cam, axis=1)
# ax = sns.heatmap(cam)
# plt.show()
# cam = cam - np.min(cam)
# cam = cam / np.max(cam)
heatmap = (cam - np.min(cam)) / (np.max(cam) - np.min(cam) + 1e-10)
print(heatmap.shape)
return heatmap
class GradCAM(BaseCAM):
def __init__(self, model, target_layer, use_cuda=False):
super(GradCAM, self).__init__(model, target_layer, use_cuda)
def get_cam_weights(self, input_tensor,
target_category,
activations,
grads):
grads_power_2 = grads ** 2
grads_power_3 = grads_power_2 * grads
sum_activations = np.sum(activations, axis=1)
eps = 0.000001
aij = grads_power_2 / (2 * grads_power_2 + sum_activations[:, None] * grads_power_3 + eps)
aij = np.where(grads != 0, aij, 0)
weights = np.maximum(grads, 0) * aij
weights = np.sum(weights, axis=1)
return weights
from pytorch_grad_cam.utils.image import preprocess_image
model = Net1()
model.load_state_dict(torch.load('./data7/G0503_02.pt'))
target_layer = model.p3_0
net = GradCAM(model, target_layer)
from settestpu import Test
from scipy.fftpack import fft
input_tensor = Test.Data[0:1, :]
input_tensor = torch.tensor(input_tensor, dtype=torch.float32)
#plt.figure(figsize=(5, 1))
output = net(input_tensor)
import scipy.io as scio
input_tensor = input_tensor.numpy().squeeze()
dataNew = "G:\\datanew.mat"
scio.savemat(dataNew, mdict={'cam': output,'data': input_tensor})