-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataset.py
182 lines (158 loc) · 7.89 KB
/
dataset.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
import random
import argparse
from abc import ABC
import numpy as np
import torch
from torch.utils.data import Dataset
import os
from initpack.seed_init import place_seed_points
from utils import *
class MetaSliceData_train:
def __init__(self, datasets, iter_n=100):
super().__init__()
self.datasets = datasets
self.dataset_n = len(datasets)
self.iter_n = iter_n
def __len__(self):
return self.iter_n
def __getitem__(self, idx):
dataset = random.sample(self.datasets, 1)[0]
return dataset.__getitem__(idx)
def meta_data(args):
tasks = [0, 1, 2, 3]
target_task = args.target_task
print('target_task is : {}'.format(target_task))
train_dataset = []
tasks.remove(target_task)
for task in tasks:
train_dataset.append(TrainLoader(task, args))
test_dataset = TestLoader(args)
meta_tr_dataset = MetaSliceData_train(train_dataset)
return meta_tr_dataset, test_dataset
class TrainLoader(Dataset, ABC):
def __init__(self, task_idx, args):
super().__init__()
self.task_idx = task_idx
self.target_task = args.target_task
self.data_dir = args.data_dir
self.T0_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.task_idx), 'T0_IMG.npy'))
self.T1_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.task_idx), 'T1_IMG.npy'))
self.T2_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.task_idx), 'T2_IMG.npy'))
self.labels = np.load(os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.task_idx), 'MASK.npy'))
self.slices = args.slices
self.shot = args.k_shot
self.imgs_num_3d = self.T0_imgs.shape[0]
self.img_slices = self.T0_imgs.shape[1]
self.max_sp = args.max_sp
def __len__(self):
return self.imgs_num_3d
def __getitem__(self, item):
idx_space = [i for i in range(self.imgs_num_3d)]
sub_idxs = random.sample(idx_space, self.shot + 1)
s_sub_idxs = sub_idxs[:self.shot]
q_sub_idx = sub_idxs[self.shot]
s_T0_imgs = self.T0_imgs[s_sub_idxs, ...]
s_T1_imgs = self.T1_imgs[s_sub_idxs, ...]
s_T2_imgs = self.T2_imgs[s_sub_idxs, ...]
s_labels = self.labels[s_sub_idxs, ...]
q_T0_imgs = self.T0_imgs[q_sub_idx, ...]
q_T1_imgs = self.T1_imgs[q_sub_idx, ...]
q_T2_imgs = self.T2_imgs[q_sub_idx, ...]
q_labels = self.labels[q_sub_idx, ...]
init_seed_list = []
for i in range(self.shot):
seed_list = []
for framid in range(self.slices):
mask = s_labels[i, framid, :, :, :] # 1 x H x W
mask = mask.reshape(-1, mask.shape[-1]) #256x256
init_seed = place_seed_points(mask, down_stride=8, max_num_sp=self.max_sp, avg_sp_area=9)
seed_list.append(init_seed.unsqueeze(0))
s_init_seed1 = torch.cat(seed_list, 0) # (slice, max_num_sp, 2)
init_seed_list.append(s_init_seed1.unsqueeze(0))
s_init_seed = torch.cat(init_seed_list, 0) #(shot, slice, max_num_sp, 2)
s_T0_imgs, s_T1_imgs, s_T2_imgs, s_labels, q_T0_imgs, q_T1_imgs, q_T2_imgs, q_labels = random_augment(s_T0_imgs,
s_T1_imgs,
s_T2_imgs,
s_labels,
q_T0_imgs,
q_T1_imgs,
q_T2_imgs,
q_labels)
sample = {
's_T0_x': to_tensor(s_T0_imgs),
's_T1_x': to_tensor(s_T1_imgs),
's_T2_x': to_tensor(s_T2_imgs),
's_y': to_tensor(s_labels),
'q_T0_x': to_tensor(q_T0_imgs),
'q_T1_x': to_tensor(q_T1_imgs),
'q_T2_x': to_tensor(q_T2_imgs),
'q_y': to_tensor(q_labels),
}
return sample, s_init_seed
class TestLoader(Dataset, ABC):
def __init__(self, args):
super().__init__()
self.target_task = args.target_task
self.data_dir = args.data_dir
self.s_T0_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.target_task), 'T0_IMG.npy'))
self.s_T1_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.target_task), 'T1_IMG.npy'))
self.s_T2_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.target_task), 'T2_IMG.npy'))
self.s_labels = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'training', str(self.target_task), 'MASK.npy'))
self.q_T0_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'testing', str(self.target_task), 'T0_IMG.npy'))
self.q_T1_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'testing', str(self.target_task), 'T1_IMG.npy'))
self.q_T2_imgs = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'testing', str(self.target_task), 'T2_IMG.npy'))
self.q_labels = np.load(
os.path.join(self.data_dir, 'Artificial_1', 'testing', str(self.target_task), 'MASK.npy'))
self.slices = args.slices
self.shot = args.k_shot
self.s_imgs_num = self.s_T0_imgs.shape[0]
self.q_imgs_num = self.q_T0_imgs.shape[0]
self.img_slices = self.q_T0_imgs.shape[1]
self.max_sp = args.max_sp
def __len__(self):
return self.q_imgs_num
def __getitem__(self, idx):
s_idx_space = [i for i in range(self.s_imgs_num)]
sub_idxs = random.sample(s_idx_space, self.shot)
s_sub_idxs = sub_idxs
q_sub_idx = idx
s_T0_imgs = self.s_T0_imgs[s_sub_idxs, ...]
s_T1_imgs = self.s_T1_imgs[s_sub_idxs, ...]
s_T2_imgs = self.s_T2_imgs[s_sub_idxs, ...]
s_labels = self.s_labels[s_sub_idxs, ...]
q_T0_imgs = self.q_T0_imgs[q_sub_idx, ...]
q_T1_imgs = self.q_T1_imgs[q_sub_idx, ...]
q_T2_imgs = self.q_T2_imgs[q_sub_idx, ...]
q_labels = self.q_labels[q_sub_idx, ...]
sample = {
's_T0_x': to_tensor(s_T0_imgs),
's_T1_x': to_tensor(s_T1_imgs),
's_T2_x': to_tensor(s_T2_imgs),
's_y': to_tensor(s_labels),
'q_T0_x': to_tensor(q_T0_imgs),
'q_T1_x': to_tensor(q_T1_imgs),
'q_T2_x': to_tensor(q_T2_imgs),
'q_y': to_tensor(q_labels),
}
init_seed_list = []
for i in range(self.shot):
seed_list = []
for framid in range(self.slices):
mask = s_labels[i, framid, :, :, :] # 1 x H x W
mask = mask.reshape(-1, mask.shape[-1]) # 256x256
init_seed = place_seed_points(mask, down_stride=8, max_num_sp=self.max_sp, avg_sp_area=9)
seed_list.append(init_seed.unsqueeze(0))
s_init_seed1 = torch.cat(seed_list, 0) # (slice, max_num_sp, 2)
init_seed_list.append(s_init_seed1.unsqueeze(0))
s_init_seed = torch.cat(init_seed_list, 0) # (shot, slice, max_num_sp, 2)
return sample, s_init_seed