-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathaugmentation.py
184 lines (143 loc) · 7.42 KB
/
augmentation.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
import os
import random
import torch
import torch.nn.functional as F
from torch import nn
from scipy.optimize import linear_sum_assignment
from einops import rearrange
import nibabel as nib
from torch.utils import data
from torch.utils.data import DataLoader
import numpy as np
from scipy.ndimage import zoom
from architecture import spatio_temporal_semantic_learning
import functions as funcs
import optimization as opt_fn
from config import opt
class clinically_credible_augmentation(nn.Module):
def __init__(self, input_shape):
super().__init__()
self.input_shape = input_shape
def data_resize(self, original_data):
data, label = original_data['cpr_volume'], original_data['label']
inp_size = data.shape
if data.shape[0] == self.input_shape[0]:
return original_data
data_zoom_factors = (
self.input_shape[0] / inp_size[0], self.input_shape[1] / inp_size[1], self.input_shape[2] / inp_size[2])
resized_data = zoom(data, data_zoom_factors, order=3)
label_zoom_factors = (self.input_shape[0] / inp_size[0])
resized_label = zoom(label, label_zoom_factors, order=1)
return {"cpr_volume": resized_data, "label": resized_label}
def data_generator(self, foreground_data, background_data):
foreground_data = self.data_resize(foreground_data)
background_data = self.data_resize(background_data)
f_data, f_label = foreground_data['cpr_volume'], foreground_data['label']
b_data, b_label = background_data['cpr_volume'], background_data['label']
ret_data = np.full((256, 64, 64), -1024, dtype=np.int32)
ret_label = f_label
b_indices = np.where(b_label == 0)[0]
ret_data[b_indices, :, :] = b_data[b_indices, :, :]
f_indices = np.where(f_label > 0)[0]
ret_data[f_indices, :, :] = f_data[f_indices, :, :]
ret_label = np.where(ret_label > 0, ((ret_label - 1) % 3) + 1, ret_label)
return {'cpr_volume': ret_data, 'label': ret_label}
def read_data(self, volumes_file, labels_file):
nii_file = nib.load(volumes_file)
affine_matrix = nii_file.affine
ret_volumes = nii_file.get_fdata()
ret_volumes = ret_volumes.transpose(2, 0, 1)
ret_volumes = np.array(ret_volumes)
ret_labels = np.loadtxt(labels_file).astype(np.int32)
return {'cpr_volume': ret_volumes, 'label': ret_labels}, affine_matrix
def write_data(self, data_info, idx, augmented_root, affine_matrix):
volumes = data_info['cpr_volume']
labels = data_info['label']
nii_image = nib.Nifti1Image(volumes, affine=affine_matrix)
nib.save(nii_image, os.path.join(augmented_root, f'volumes/gen_{idx}.nii'))
labels_str = '\n'.join(map(str, labels))
with open(os.path.join(augmented_root, f'labels/gen_{idx}.txt'), 'w') as file:
file.write(labels_str)
return
def forward(self, generated_num, original_root=r'original_data_root', augmented_root=r'augmented_data_root'):
volumes_root = os.path.join(original_root, 'volumes/')
labels_root = os.path.join(original_root, 'labels/')
volumes_file_list = os.listdir(volumes_root)
volumes_file_list = sorted(volumes_file_list)
labels_file_list = os.listdir(labels_root)
labels_file_list = sorted(labels_file_list)
file_total = len(volumes_file_list)
for i in range(generated_num):
selected_f_idx, selected_d_idx = random.randint(0, file_total - 1), random.randint(0, file_total - 1)
volumes_f_file = os.path.join(volumes_root, volumes_file_list[selected_f_idx])
labels_f_file = os.path.join(labels_root, labels_file_list[selected_f_idx])
volumes_d_file = os.path.join(volumes_root, volumes_file_list[selected_d_idx])
labels_d_file = os.path.join(labels_root, labels_file_list[selected_d_idx])
ori_f_data, affine_matrix = self.read_data(volumes_f_file, labels_f_file)
ori_b_data, affine_matrix = self.read_data(volumes_d_file, labels_d_file)
gen_data = self.data_generator(ori_f_data, ori_b_data)
self.write_data(gen_data, i, augmented_root, affine_matrix)
return
class cubic_sequence_data(data.Dataset):
def __init__(self, dataset_root, pattern='training', train_ratio=0.8, input_shape=[256,64,64], window=[300, 900]):
self.volumes_root = os.path.join(dataset_root, 'volumes/')
self.labels_root = os.path.join(dataset_root, 'labels/')
self.input_shape, self.window = input_shape, [window[0] - window[1] / 2, window[0] + window[1] / 2]
self.volumes_file_list = os.listdir(self.volumes_root)
self.volumes_file_list = sorted(self.volumes_file_list)
self.labels_file_list = os.listdir(self.labels_root)
self.labels_file_list = sorted(self.labels_file_list)
self.file_total = len(self.volumes_file_list)
if pattern == 'training':
self.data_start, self.data_end = 0, int(self.file_total * train_ratio)
else :
self.data_start, self.data_end = int(self.file_total * train_ratio), self.file_total
self.length = self.data_end - self.data_start
return
def read_data(self, volumes_file, labels_file):
nii_file = nib.load(volumes_file)
ret_volumes = nii_file.get_fdata()
if ret_volumes.shape[0] == ret_volumes.shape[1]:
ret_volumes = ret_volumes.transpose(2, 0, 1)
ret_volumes = np.array(ret_volumes)
ret_labels = np.loadtxt(labels_file).astype(np.int32)
return ret_volumes, ret_labels
def detection_targets(self, labels_data):
boxes, labels = [], []
start, label, length, last = None, 0, self.input_shape[0], -1
for i in range(labels_data.shape[0]):
if start is not None:
if labels_data[i] != last:
boxes.append([(start + 1) / length, min((i + 1) / length, 1.0)])
labels.append(label - 1)
if labels_data[i] != 0:
start, label, last = i, labels_data[i], labels_data[i]
else: start, label, last = None, 0, -1
else:
continue
else:
if labels_data[i] == 0:
start, label, last = None, 0, -1
else:
start, label, last = i, labels_data[i], labels_data[i]
if start is not None:
boxes.append([(start + 1) / length, 1.0])
labels.append(label - 1)
labels = torch.tensor(labels, dtype=torch.int64)
boxes = torch.tensor(boxes, dtype=torch.torch.float32)
return {"labels": labels, "boxes": boxes}
def __getitem__(self, index):
volumes_file = os.path.join(self.volumes_root, self.volumes_file_list[index])
labels_file = os.path.join(self.labels_root, self.labels_file_list[index])
ret_volumes, ret_labels = self.read_data(volumes_file, labels_file)
ret_volumes = funcs.normalize_ct_data(ret_volumes, hu_min=self.window[0], hu_max=self.window[1])
return {'image': torch.tensor(ret_volumes,dtype=torch.float32), 'target': self.detection_targets(ret_labels)}
def __len__(self):
return self.length
def collate_fn(batch):
images, targets = [], []
for item in batch:
images.append(item['image'])
targets.append(item['target'])
images = torch.stack(images, dim=0)
return images, targets