-
Notifications
You must be signed in to change notification settings - Fork 0
/
lrp_tb.py
140 lines (117 loc) · 5.19 KB
/
lrp_tb.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
import imp
import matplotlib.pyplot as plt
import numpy as np
import os
import innvestigate
import innvestigate.utils
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation, Flatten
from keras.layers.convolutional import Conv2D, MaxPooling2D, ZeroPadding2D
from keras.layers.normalization import BatchNormalization
from keras.applications.vgg16 import preprocess_input
import keras
import torch
from scipy.io import savemat
from PIL import Image
base_dir = os.path.dirname(__file__)
utils = imp.load_source("utils", os.path.join(base_dir, "utils_lrp.py"))
def alexnet_model(img_shape=(224, 224, 3), n_classes=50):
# Initialize model
alexnet = Sequential()
# Layer 1
alexnet.add(Conv2D(64, (11, 11), strides=(4,4), input_shape=img_shape, padding='valid'))
alexnet.add(Activation('relu'))
alexnet.add(MaxPooling2D(pool_size=(3,3), strides=(2,2)))
# Layer 2
alexnet.add(ZeroPadding2D((2, 2)))
alexnet.add(Conv2D(192, (5, 5), strides=(1,1), padding='valid'))
alexnet.add(Activation('relu'))
alexnet.add(MaxPooling2D(pool_size=(3,3), strides=(2,2)))
# Layer 3
alexnet.add(ZeroPadding2D((1, 1)))
alexnet.add(Conv2D(384, (3, 3), strides=(1,1), padding='valid'))
alexnet.add(Activation('relu'))
# Layer 4
alexnet.add(ZeroPadding2D((1, 1)))
alexnet.add(Conv2D(256, (3, 3), strides=(1,1), padding='valid'))
alexnet.add(Activation('relu'))
# Layer 5
alexnet.add(ZeroPadding2D((1, 1)))
alexnet.add(Conv2D(256, (3, 3), strides=(1,1), padding='valid'))
alexnet.add(Activation('relu'))
alexnet.add(MaxPooling2D(pool_size=(3,3), strides=(2,2)))
# Layer 6
alexnet.add(Flatten())
alexnet.add(Dropout(0.5))
alexnet.add(Dense(4096))
alexnet.add(Activation('relu'))
# Layer 7
alexnet.add(Dropout(0.5))
alexnet.add(Dense(4096))
alexnet.add(Activation('relu'))
# Layer 8
alexnet.add(Dense(n_classes))
alexnet.add(Activation('softmax'))
return alexnet
if __name__ == "__main__":
data_dir = "./data/CASIA_WebFace_5000"
data_name = 'test_crop'
epoch = '40'
img_dir = os.path.join(data_dir, data_name)
log_dir = f'./logs/CASIA_WebFace_5000_0.2/[alex]_[0.001]_[0.5]_[train_crop_mouth0.2_{epoch}]'
task = f'train_crop_mouth0.2_{epoch}'
lrp_dir = './lrp_result_new/' + task
if not os.path.exists(lrp_dir):
os.makedirs(lrp_dir)
pytorch_pretrain_ckpt = os.path.join(log_dir, '200.pth')
pytorch_para_dict = torch.load(pytorch_pretrain_ckpt)
keras_model = keras.models.Sequential([
keras.layers.Conv2D(filters=64, kernel_size=(11,11), strides=(4,4), activation='relu', input_shape=(224,224,3)),
keras.layers.MaxPool2D(pool_size=(3,3), strides=(2,2)),
keras.layers.Conv2D(filters=192, kernel_size=(5,5), strides=(1,1), activation='relu', padding="valid"),
keras.layers.MaxPool2D(pool_size=(3,3), strides=(2,2)),
keras.layers.Conv2D(filters=384, kernel_size=(3,3), strides=(1,1), activation='relu', padding="valid"),
keras.layers.Conv2D(filters=256, kernel_size=(3,3), strides=(1,1), activation='relu', padding="valid"),
keras.layers.Conv2D(filters=256, kernel_size=(3,3), strides=(1,1), activation='relu', padding="valid"),
keras.layers.MaxPool2D(pool_size=(3,3), strides=(2,2)),
keras.layers.Flatten(),
keras.layers.Dropout(0.5),
keras.layers.Dense(4096, activation='relu'),
keras.layers.Dropout(0.5),
keras.layers.Dense(4096, activation='relu'),
keras.layers.Dense(50, activation='softmax')
])
keras_model = alexnet_model()
keras_model.summary()
keras_weights = keras_model.get_weights()
layer_list = [
'features.0', 'features.3',
'features.6', 'features.8', 'features.10',
'classifier.1', 'classifier.4', 'classifier.6'
]
for i, layer in enumerate(layer_list):
ind = 2 * i
print(f'\n[+] weights index: {ind}, layer name: {layer}')
weight_key = layer + '.weight'
bias_key = layer + '.bias'
print(f'[-] pytorch weight {pytorch_para_dict[weight_key].shape} and keras weight {keras_weights[ind].shape}')
keras_weights[ind] = np.transpose(pytorch_para_dict[weight_key].cpu().numpy())
print(f'[-] pytorch bias {pytorch_para_dict[bias_key].shape} and keras bias {keras_weights[ind+1].shape}')
keras_weights[ind+1] = np.transpose(pytorch_para_dict[bias_key].cpu().numpy())
keras_model.set_weights(keras_weights)
keras_model = innvestigate.utils.model_wo_softmax(keras_model)
lrp_tag = 'lrp.sequential_preset_a_flat'
analyzer = innvestigate.create_analyzer(
lrp_tag, keras_model,
neuron_selection_mode="index"
)
image_path = '/home/sda1/Jinge/Attention_analysis/data/CASIA_WebFace_5000/test_crop/0005095/071.jpg'
image = utils.load_image(image_path, 224)
x = preprocess_input(image[None])
result = analyzer.analyze(x, neuron_selection=0)
# Aggregate along color channels and normalize to [-1, 1]
a = result.sum(axis=np.argmax(np.asarray(result.shape) == 3))
a /= np.max(np.abs(a))
# R_dict.update({'f' + img_name: a[0]})
plt.imshow(a[0], cmap="seismic", clim=(-1, 1))
plt.title(lrp_tag)