Newer
Older
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
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
# -*- coding: utf-8 -*-
"""
Created on Wed Mar 7 16:42:15 2018
This file is for testing the already trained active learning model
@author: s161488
"""
import tensorflow as tf
from data_utils.prepare_data import padding_training_data, prepare_test_data
from models.inference import ResNet_V2_DMNN
import os
import numpy as np
import scipy.io as sio
data_home = "/home/blia/Region_Active_Learning/DATA/Data/" # NOTE, NEED TO BE MANUALLY DEFINED
def running_test_for_all_acquisition_steps(path_input, version_use, start_step, pool_or_test):
"""This function evaluates the experiments for multiple acquisition steps
Args:
path_input: "Method_B_Stage_1_Version_0", the parent directory of each acquisition step ckpt dir
version_use: int
start_step: int, default 0, where to start the evaluation
pool_or_test: whether I am evaluating the experiments on the pool dataset or test dataset, string,
"pool" or "test"
Note:
the path_input and version_use argument can be omitted depends on the place that you save your experiments
"""
data_kind = ['Test_Data_A/', 'Test_Data_B/']
if pool_or_test is "test":
data_name = ['glanddata_testa.npy', 'glanddata_testb.npy']
else:
data_name = ["glanddata.npy"]
path_mom = os.path.join('/scratch/blia/Act_Learn_Desperate_V%d/' % version_use, path_input)
path_all_cousin = np.load(os.path.join(path_mom, 'total_select_folder.npy'))[start_step:]
if pool_or_test is "test":
test_data_init_path = os.path.join(path_mom, 'Test_Data')
else:
test_data_init_path = os.path.join(path_mom, 'Pool_Data')
for path_single_cousin in path_all_cousin:
ckpt_to_read = path_mom + path_single_cousin.strip().split(path_input)[-1]
# ckpt_to_read = path_single_cousin
for data_path_single, data_name_single in zip(data_kind, data_name):
single_folder_name = path_single_cousin.strip().split('/')[-2]
test_data_statistics_dir = os.path.join(test_data_init_path, single_folder_name)
test_data_statistics_dir = os.path.join(test_data_statistics_dir, data_path_single)
if not os.path.exists(test_data_statistics_dir):
os.makedirs(test_data_statistics_dir)
test_data_path = os.path.join(data_home, data_name_single)
test(test_data_statistics_dir, ckpt_to_read, test_data_path, save_stat=True)
# save_image_to_mat(test_data_statistics_dir, test_data_path)
def running_test_for_single_acquisition_step(model_dir):
"""This function saves the prediction for a single acquisition step
It performs the prediction for Group A and Group B images separately
model_dir: the directory that saves the model ckpt
Note:
the data_home needs to be manually determined!
"""
data_kind = ['Test_Data_A/', 'Test_Data_B/']
data_name = ['glanddata_testa.npy', 'glanddata_testb.npy']
tds = model_dir.strip().split('/FE')[0] + "/Test_Data/%s/" % model_dir.strip().split('/')[-2]
if not os.path.exists(tds):
os.makedirs(tds)
print("the test directory----", tds)
for data_path_single, data_name_single in zip(data_kind, data_name):
test_data_statistics_dir = os.path.join(tds, data_path_single)
if not os.path.exists(test_data_statistics_dir):
os.makedirs(test_data_statistics_dir)
test_data_path = os.path.join(data_home, data_name_single)
test(test_data_statistics_dir, model_dir, test_data_path, save_stat=True)
# save_image_to_mat(test_data_statistics_dir, test_data_path)
def test(test_data_statistics_dir, ckpt_dir, test_data_path, save_stat=True):
"""This function evaluates the performance of the model
Args:
test_data_statistics_dir: the tds_dir that is used for saving data
ckpt_dir: the directory that saves the model ckpt
test_data_path: the path that saves the test data. because there are gland A and gland B
save_stat: bool, whether save the fb probability and ed probability, default to be True
"""
# --------Here lots of parameters need to be set------Or maybe we could set it in the configuration file-----#
batch_size = 1
targ_height_npy = 528 # this is for padding images
targ_width_npy = 784 # this is for padding images
ckpt_dir = ckpt_dir
image_c = 3
MOVING_AVERAGE_DECAY = 0.999
check_effect_of_dropout_iter = False
# if check_effect_of_dropout_iter is True, then the goal is to find a good number of dropout times which can
# achieve good segmentation accuracy
if check_effect_of_dropout_iter is True:
sample_size_space = [1, 5, 10, 15, 20, 25, 30, 40, 50]
else:
sample_size_space = [30]
with tf.Graph().as_default():
# The placeholder below is for extracting the input for the network #####
images_train = tf.placeholder(tf.float32, [batch_size, targ_height_npy, targ_width_npy, image_c])
instance_labels_train = tf.placeholder(tf.int64, [batch_size, targ_height_npy, targ_width_npy])
edges_labels_train = tf.placeholder(tf.int64, [batch_size, targ_height_npy, targ_width_npy])
phase_train = tf.placeholder(tf.bool, shape=None, name="training_state")
dropout_phase = tf.placeholder(tf.bool, shape=None, name="dropout_state")
# -----------------Here is for preparing the dataset for training, pooling and validation---------###
x_val_group = prepare_test_data(test_data_path)
x_image_val, y_label_val, y_edge_val = padding_training_data(x_val_group[0], x_val_group[1], x_val_group[2],
targ_height_npy,
targ_width_npy)
print("The real validation data is actually the test data from group B %d" % np.shape(x_image_val)[0])
# ------------------------------Here is for build up the network-----------------------------###
fb_logits, ed_logits = ResNet_V2_DMNN(images=images_train, training_state=phase_train,
dropout_state=dropout_phase, Num_Classes=2)
edge_prob = tf.nn.softmax(tf.add_n(ed_logits))
fb_prob = tf.nn.softmax(tf.add_n(fb_logits))
var_train = tf.trainable_variables()
variable_averages = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY)
variable_averages.apply(var_train)
variables_to_restore = variable_averages.variables_to_restore(tf.moving_average_variables())
saver = tf.train.Saver(variables_to_restore)
print(" =====================================================")
with tf.Session() as sess:
ckpt_all = tf.train.get_checkpoint_state(ckpt_dir)
ckpt_use = ckpt_dir + '/' + ckpt_all.model_checkpoint_path.strip().split('/')[-1]
if os.path.isfile(ckpt_use + '.index'):
saver.restore(sess, ckpt_use)
print("restore parameter from", ckpt_use)
ed_prob_tot = []
fb_prob_tot = []
fb_bald_tot = []
num_image = np.shape(x_image_val)[0]
print("fb loss, fb acc, ed loss, ed accu")
tot_stat = np.zeros([num_image, 2])
for single_image in range(num_image):
feed_dict_op = {images_train: np.expand_dims(x_image_val[single_image], 0),
instance_labels_train: np.expand_dims(y_label_val[single_image], 0),
edges_labels_train: np.expand_dims(y_edge_val[single_image], 0),
phase_train: False,
dropout_phase: True}
fb_prob_per_image_tot = []
ed_prob_per_image_tot = []
fb_bald_per_image_tot = []
for sample_iter, single_sample_size in enumerate(sample_size_space):
fb_prob_per_image = []
ed_prob_per_image = []
fb_bald_per_image = []
for single_sample in range(single_sample_size):
_fb_prob, _ed_prob = sess.run([fb_prob, edge_prob], feed_dict=feed_dict_op)
fb_prob_per_image.append(_fb_prob)
ed_prob_per_image.append(_ed_prob)
single_fb_bald = _fb_prob * np.log(_fb_prob + 1e-08)
# single_ed_bald = _ed_prob*np.log(_ed_prob+1e-08)
fb_bald_per_image.append(single_fb_bald)
fb_prob_per_im_avg = np.mean(fb_prob_per_image, axis=0)
ed_prob_per_im_avg = np.mean(ed_prob_per_image, axis=0)
fb_bald_per_im_avg = np.mean(fb_bald_per_image, axis=0)
fb_prob_per_image_tot.append(fb_prob_per_im_avg)
ed_prob_per_image_tot.append(ed_prob_per_im_avg)
fb_bald_per_image_tot.append(fb_bald_per_im_avg)
if check_effect_of_dropout_iter is False:
fb_prob_tot.append(fb_prob_per_image_tot)
ed_prob_tot.append(ed_prob_per_image_tot)
fb_bald_tot.append(fb_bald_per_image_tot)
# fb_prob_var_tot.append(fb_prob_per_image_var_tot)
# ed_prob_var_tot.append(ed_prob_per_image_var_tot)
if save_stat is False:
pass
else:
np.save(os.path.join(test_data_statistics_dir, 'fbprob'), fb_prob_tot)
np.save(os.path.join(test_data_statistics_dir, 'edprob'), ed_prob_tot)
np.save(os.path.join(test_data_statistics_dir, 'fbbald'), fb_bald_tot)
def save_image_to_mat(test_data_statistics_dir, test_data_path):
""""This function saves the npy stat into mat stat in order to evaluate the segmentation accuracy in Matlab"""
targ_height_npy = 528 # this is for padding images
targ_width_npy = 784 # this is for padding images
x_im_group = prepare_test_data(test_data_path)
x_image_val, y_label_val, y_edge_val = padding_training_data(x_im_group[0], x_im_group[1], x_im_group[2],
targ_height_npy, targ_width_npy)
fb_prob = np.squeeze(np.squeeze(np.load(os.path.join(test_data_statistics_dir, 'fbprob.npy')), 1), 1)
ed_prob = np.squeeze(np.squeeze(np.load(os.path.join(test_data_statistics_dir, 'edprob.npy')), 1), 1)
mask_final_a = []
ed_prob_cut = []
fb_prob_cut = []
for index in range(np.shape(ed_prob)[0]):
sele_index = np.where(np.mean(x_image_val[index], -1) != 0)
ed_prob_single = ed_prob[index]
fb_prob_single = fb_prob[index]
mask_gt_single = y_label_val[index][np.min(sele_index[0]):np.max(sele_index[0] + 1),
np.min(sele_index[1]):np.max(sele_index[1] + 1)]
ed_prob_cut.append(ed_prob_single[np.min(sele_index[0]):np.max(sele_index[0] + 1),
np.min(sele_index[1]):np.max(sele_index[1] + 1), :])
fb_prob_cut.append(fb_prob_single[np.min(sele_index[0]):np.max(sele_index[0] + 1),
np.min(sele_index[1]):np.max(sele_index[1] + 1), :])
mask_final_a.append(mask_gt_single != 0)
sio.savemat(os.path.join(test_data_statistics_dir, 'edge_pred.mat'), {'pred': ed_prob_cut})
sio.savemat(os.path.join(test_data_statistics_dir, 'fb_pred.mat'), {'pred': fb_prob_cut})
sio.savemat(os.path.join(test_data_statistics_dir, 'mask.mat'), {'ground_truth': mask_final_a})
os.remove(os.path.join(test_data_statistics_dir, 'fbprob.npy'))
os.remove(os.path.join(test_data_statistics_dir, 'edprob.npy'))