Newer
Older
# -*- coding: utf-8 -*-
"""
Created on Wed May 16 14:27:21 2018
This script is for preparing the input for the active learning, we need to have training data, pool data,
validation data.
@author: s161488
"""
import numpy as np
import tensorflow as tf
import cv2
import matplotlib.pyplot as plt
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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
def prepare_train_data(path, select_benign_train, select_mali_train):
"""
Args:
path: the path where the data is saved
select_benign_train: a list of selected benign images
select_mali_train: a list of selected malignant images
Ops:
First, the images, labels. edges, im_index, cls_index can be extracted from the np.load
images: shape [85, im_h, im_w, 3]
labels: shape [85, im_h, im_w, 1]
im_index: shape [85]
cls_index: shape [85]
Start_Point will determine how many images are initialized as training image
Output:
X_train, Y_train
X_pool, Y_pool
X_val, Y_val
"""
data_set = np.load(path, allow_pickle=True).item()
images = data_set['image']
labels = data_set['label']
edges = data_set['edge']
imageindex = data_set['ImageIndex']
classindex = data_set['ClassIndex']
benign_index = np.where(np.array(classindex) == 1)
mali_index = np.where(np.array(classindex) == 2)
choose_index_tr = np.concatenate([benign_index[0][select_benign_train], mali_index[0][select_mali_train]], axis=0)
benign_index_left = np.delete(range(np.shape(benign_index[0])[0]), select_benign_train)
mali_index_left = np.delete(range(np.shape(mali_index[0])[0]), select_mali_train)
choose_index_pl = np.concatenate([benign_index[0][benign_index_left[:27]], mali_index[0][mali_index_left[:38]]],
axis=0)
choose_index_val = np.concatenate([benign_index[0][benign_index_left[-5:]], mali_index[0][mali_index_left[-5:]]],
axis=0)
data_train = extract_diff_data(images, labels, edges, imageindex, classindex, choose_index_tr)
data_pl = extract_diff_data(images, labels, edges, imageindex, classindex, choose_index_pl)
data_val = extract_diff_data(images, labels, edges, imageindex, classindex, choose_index_val)
return data_train, data_pl, data_val
def prepare_pool_data(path, aug=False):
data_set = np.load(path, allow_pickle=True).item()
images = data_set['image']
labels = data_set['label']
edges = data_set['edge']
imageindex = data_set['ImageIndex']
classindex = data_set['ClassIndex']
select_benign_train = [0, 1, 2, 3, 4]
select_mali_train = [2, 4, 5, 6, 7]
benign_index = np.where(np.array(classindex) == 1)
mali_index = np.where(np.array(classindex) == 2)
benign_index_left = np.delete(range(np.shape(benign_index[0])[0]), select_benign_train)
mali_index_left = np.delete(range(np.shape(mali_index[0])[0]), select_mali_train)
choose_index_pl = np.concatenate([benign_index[0][benign_index_left[:27]], mali_index[0][mali_index_left[:38]]],
axis=0)
data_pl = extract_diff_data(images, labels, edges, imageindex, classindex, choose_index_pl)
if aug is True:
targ_height_npy = 528 # this is for padding images
targ_width_npy = 784 # this is for padding images
x_image_val, y_label_val, y_edge_val = padding_training_data(data_pl[0], data_pl[1],
data_pl[2], targ_height_npy,
targ_width_npy)
data_pl = [x_image_val, y_label_val, y_edge_val]
return data_pl
def prepare_skin_data(path, num_tr, combine=True):
"""
choose_index_tr: worst 16+best 16
or middle 32
this num_tr should be 1/2*total_number_of_training_images_at_inital_step
I have tried it for 32, then I am going to check 16
"""
val_num_im = 96
tot_numeric_index = np.arange(900)
if combine is True:
tr_select_numeric_index = np.concatenate([tot_numeric_index[:num_tr], tot_numeric_index[-num_tr:]], axis=0)
else:
tr_select_numeric_index = tot_numeric_index[340:(340 + num_tr * 2)]
val_select_numeric_index = tot_numeric_index[500:(500 + val_num_im)]
pool_numeric_index = np.delete(tot_numeric_index,
np.concatenate([tr_select_numeric_index, val_select_numeric_index], axis=0))
im_seg_score = np.load('/home/s161488/Exp_Stat/Skin_Lesion/init_segment_score.npy')
sorted_index = np.argsort(im_seg_score)
data_set = np.load(path, encoding='latin1').item()
images = np.array(data_set['image'])
labels = np.array(data_set['label'])
edges = np.array(data_set['edge'])
labels = np.expand_dims(labels, axis=-1)
edges = np.expand_dims(edges, axis=-1)
tr_select_image_index = np.sort(sorted_index[tr_select_numeric_index])
val_select_image_index = np.sort(sorted_index[val_select_numeric_index])
pl_select_image_index = np.sort(sorted_index[pool_numeric_index])
imindex = np.arange(np.shape(images)[0])
clsindex = np.ones(np.shape(images)[0])
data_tr = extract_diff_data(images, labels, edges, imindex, clsindex, tr_select_image_index)
data_pl = extract_diff_data(images, labels, edges, imindex, clsindex, pl_select_image_index)
data_val = extract_diff_data(images, labels, edges, imindex, clsindex, val_select_image_index)
return data_tr[:3], data_pl[:3], data_val[:3]
def prepare_test_data(path):
if "_test" not in path:
print("-------I am loading the data from pool set------")
return prepare_pool_data(path)
data_set = np.load(path, allow_pickle=True).item()
images = data_set['image']
labels = data_set['label']
edges = data_set['edge']
imageindex = data_set['ImageIndex']
classindex = data_set['ClassIndex']
return images, labels, edges, imageindex, classindex
def generate_batch(x_image_tr, y_label_tr, y_edge_tr, y_binary_mask_tr, batch_index, batch_size):
im_group = [x_image_tr, y_label_tr, y_edge_tr, y_binary_mask_tr]
im_batch = []
for single_im in im_group:
_im_batch = single_im[batch_index:(batch_size + batch_index)]
im_batch.append(_im_batch)
batch_index = batch_index + batch_size
return im_batch[0], im_batch[1], im_batch[2], im_batch[3], batch_index
def padding_training_data(x_image, y_label, y_edge, target_height, target_width):
"""Each image has different size, so I need to pad it with zeros to make sure each image has the same size.
Then I can perform random crop, rotation and other augmentation on per batch instead of per image
"""
x_im_pad, y_la_pad, y_ed_pad = [], [], []
num_image = np.shape(x_image)[0]
for i in range(num_image):
image_pad, label_pad, edge_pad = padding_zeros(x_image[i], y_label[i], y_edge[i], target_height, target_width)
x_im_pad.append(image_pad)
y_la_pad.append(label_pad)
y_ed_pad.append(edge_pad)
x_im_pad = np.reshape(x_im_pad, [num_image, target_height, target_width, 3])
y_la_pad = np.reshape(y_la_pad, [num_image, target_height, target_width, 1])
y_ed_pad = np.reshape(y_ed_pad, [num_image, target_height, target_width, 1])
return x_im_pad, y_la_pad, y_ed_pad
def padding_zeros(image, label, edge, target_height, target_width):
im_h, im_w, _ = np.shape(image)
delta_w = target_width - im_w
delta_h = target_height - im_h
top, bottom = delta_h // 2, delta_h - delta_h // 2
left, right = delta_w // 2, delta_w - delta_w // 2
image_pad = np.pad(image, ((top, bottom), (left, right), (0, 0)), mode='constant')
label_pad = np.pad(label, ((top, bottom), (left, right)), mode='constant')
edge_pad = np.pad(edge, ((top, bottom), (left, right)), mode='constant')
return image_pad, label_pad, edge_pad
def extract_diff_data(image, label, edge, im_index, cls_index, choose_index):
new_data = [[] for _ in range(5)]
old_data = [image, label, edge, im_index, cls_index]
for i in choose_index:
for single_new, single_old in zip(new_data, old_data):
single_new.append(single_old[i])
return new_data[0], new_data[1], new_data[2], new_data[3], new_data[4]
def aug_train_data(image, label, edge, binary_mask, batch_size, aug, imshape):
"""This function is used for performing data augmentation.
image: placeholder. shape: [Batch_Size, im_h, im_w, 3], tf.float32
label: placeholder. shape: [Batch_Size, im_h, im_w, 1], tf.int64
edge: placeholder. shape: [Batch_Size, im_h, im_w, 1], tf.int64
binary_mask: placeholder. shape: [Batch_Size, im_h, im_w, 1], tf.int64
aug: bool
imshape: [targ_h, targ_w, ch]
Outputs:
image: [Batch_Size, targ_h, targ_w, 3]
label: [Batch_Size, targ_h, targ_w, 1]
edge: [Batch_Size, targ_h, targ_w, 1]
binary_mask: [Batch_Size, targ_h, targ_w, 1]
"""
image = tf.cast(image, tf.int64)
bigmatrix = tf.concat([image, label, edge, binary_mask], axis=3)
target_height = imshape[0].astype('int32')
target_width = imshape[1].astype('int32')
if aug is True:
bigmatrix_crop = tf.random_crop(bigmatrix, size=[batch_size, target_height, target_width, 6])
bigmatrix_crop = tf.cond(tf.less_equal(tf.reduce_sum(bigmatrix_crop[:, :, :, 5]), 10),
lambda: tf.image.resize_image_with_crop_or_pad(bigmatrix, target_height, target_width),
lambda: bigmatrix_crop)
# instead of judging by label, should do it by the binary mask!
k = tf.random_uniform(shape=[batch_size], minval=0, maxval=6.5, dtype=tf.float32)
bigmatrix_rot = tf.contrib.image.rotate(bigmatrix_crop, angles=k)
image_aug = tf.cast(bigmatrix_rot[:, :, :, 0:3], tf.float32)
label_aug = bigmatrix_rot[:, :, :, 3]
edge_aug = bigmatrix_rot[:, :, :, 4]
binary_mask_aug = bigmatrix_rot[:, :, :, 5]
else:
bigmatrix_rot = tf.image.resize_image_with_crop_or_pad(bigmatrix, target_height, target_width)
image_aug = tf.cast(tf.cast(bigmatrix_rot[:, :, :, 0:3], tf.uint8), tf.float32)
label_aug = tf.cast(bigmatrix_rot[:, :, :, 3], tf.int64)
edge_aug = tf.cast(bigmatrix_rot[:, :, :, 4], tf.int64)
binary_mask_aug = tf.cast(bigmatrix_rot[:, :, :, 5], tf.int64)
return image_aug, tf.expand_dims(label_aug, -1), tf.expand_dims(edge_aug, -1), tf.expand_dims(binary_mask_aug, -1)
def collect_test_data(resize=True):
test_a_path = path_mom + "/Data/glanddata_testa.npy"
test_b_path = path_mom + "/Data/glanddata_testb.npy"
image_tot, label_tot = [], []
target_height, target_width = 528, 784
for single_path in [test_a_path, test_b_path]:
data_set = np.load(single_path, allow_pickle=True).item()
images = data_set['image']
y_label_pl = data_set['label']
y_edge_pl = data_set['edge']
x_image_val = []
y_label_val = []
if resize is True:
for single_im, single_label in zip(images, y_label_pl):
for _im_, _path_ in zip([single_im, single_label], [x_image_val, y_label_val]):
_im_ = cv2.resize(_im_, dsize=(784, 528), interpolation=cv2.INTER_CUBIC)
_path_.append(_im_)
else:
x_image_val, y_label_val, y_edge_val = padding_training_data(images, y_label_pl, y_edge_pl, target_height,
target_width)
image_tot.append(x_image_val)
label_tot.append(y_label_val)
image_tot = np.concatenate([image_tot[0], image_tot[1]], axis=0)
label_tot = np.concatenate([label_tot[0], label_tot[1]], axis=0)
print("The shape of the test images", np.shape(image_tot))
return image_tot, label_tot
def save_im():
im_tot, la_tot = collect_test_data()
rand_value = np.random.choice(np.arange(len(im_tot)), 3, replace=False)
for i in rand_value:
fig = plt.figure(figsize=(10, 4))
im_ = im_tot[i]
la_ = la_tot[i]
la_judge = (la_ != 0)
for iterr, single_im in enumerate([im_, la_, la_judge]):
ax = fig.add_subplot(1, 3, iterr + 1)
ax.imshow(single_im)
plt.savefig('/home/blia/im_%d.pdf' % i, pad_inches=0, box_inches='tight')