-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataloading.py
334 lines (231 loc) · 11.1 KB
/
dataloading.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
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
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
import h5py as h5
import numpy as np
import torch
import torch.nn as nn
from torch.utils.data import Dataset
import glob
from get_affs import GetAffs
class DISCoDataset(Dataset):
def __init__(self, path, device, dtype, mode='disco',usedata=None,
sec_length=100, sec_size=128, n_secs=10):
"""
Args:
"""
self.dtype = dtype
self.device = device
self.sec_length = sec_length
self.sec_size = sec_size
self.n_secs = n_secs
self.path = path
if usedata is not None:
training_data_ = []
self.test_set = []
training_ = glob.glob(path+'neurofinder.'+usedata+'*.h5')
test_ = glob.glob(path+'neurofinder.'+usedata+'*.test.h5')
training_data_ += training_
self.test_set += test_
else:
training_data_ = glob.glob(path+'neurofinder.*.h5')
self.test_set = glob.glob(path+'neurofinder.*.test.h5')
self.training_data = [item for item in training_data_ if item not in self.test_set]
del training_data_
self.mode = mode
videos = []
names = []
labels = []
segs = []
summs = []
for data in self.training_data:
print('loading ' + data)
name = data[len(path):-3]
with h5.File(data,'r') as in_file:
video = in_file['video'][...].astype(np.int16)
with h5.File(path+'BF_labels.h5','r') as l_file:
label = l_file[name][...]
with h5.File(path+'gt_segmentations.h5','r') as seg_file:
seg = seg_file[name][...]
with h5.File(path+'summary_images.h5','r') as s_file:
summ = s_file[name][...]
names.append(name)
videos.append(video)
labels.append(label[None,...])
segs.append(seg)
summs.append(summ[None,...])
self.data = {'videos' : videos, 'labels' : labels,
'names' : names, 'summary' : summs, 'segmentations' : segs}
self.data_length = len(names)
self.predict = False
offsets_affs = [[-1, 0], [0, -1], [-5, 0], [0, -5], [-5, -5]]
self.GA = GetAffs(offsets_affs,self.dtype,self.device)
def fetch_test_data(self,idx):
# test set
videos = []
names = []
summs = []
data = self.test_set[idx]
name = data[len(self.path):-3]
names.append(name)
with h5.File(data,'r') as in_file:
video = in_file['video'][...].astype(np.int16)
videos.append(video)
summ = np.mean(video,axis=0)
summs.append(summ[None,...])
self.data = {'videos' : videos,
'names' : names, 'summary' : summs}
self.data_length = 1
return()
def fetch_train_data(self,idx):
# test set
videos = []
names = []
summs = []
data = self.training_data[idx]
name = data[len(self.path):-3]
names.append(name)
with h5.File(data,'r') as in_file:
video = in_file['video'][...].astype(np.int16)
videos.append(video)
with h5.File(self.path+'summary_images.h5','r') as s_file:
summ = s_file[name][...]
summs.append(summ[None,...])
self.data = {'videos' : videos,
'names' : names, 'summary' : summs}
self.data_length = 1
return()
def get_transform_params(self, seg_o):
# seg_o: np array, 1 x X x Y
# no transformations for prediction
hflip = False
vflip = False
rots = 0
maxpool_size = 6
x_start = 0
x_stop = seg_o.size(1)
y_start = 0
y_stop = seg_o.size(2)
if self.predict == False:
# random transformations for training
# vflip
if np.random.random() > 0.5:
vflip = True
# hflip
if np.random.random() > 0.5:
hflip = True
# rotate
rots = np.random.choice(4,1)[0]
# temporal length for maxpooling
maxpool_size = np.random.choice(np.arange(3,10),1)[0]
# random crop
poss_start = 0
x_start = np.random.choice(np.arange(poss_start,seg_o.size(1)-self.sec_size),1)[0]
x_stop = x_start + self.sec_size
y_start = np.random.choice(seg_o.size(2)-self.sec_size,1)[0]
y_stop = y_start + self.sec_size
seg = seg_o[:,x_start:x_stop,y_start:y_stop]
u = torch.unique(seg)
while len(u) < 1:
x_start = np.random.choice(seg_o.size(1)-self.sec_size,1)[0]
x_stop = x_start + self.sec_size
y_start = np.random.choice(seg_o.size(2)-self.sec_size,1)[0]
y_stop = y_start + self.sec_size
seg = seg_o[:,x_start:x_stop,y_start:y_stop]
u = torch.unique(seg)
return([hflip,vflip,rots,maxpool_size,x_start,x_stop,y_start,y_stop])
def transform_summlike(self,summ,transform_params,norm=True,comp_affs=False):
hflip,vflip,rots,maxpool_size,x_start,x_stop,y_start,y_stop = transform_params
summ = summ[:,x_start:x_stop,y_start:y_stop]
if vflip == True:
summ = summ.flip(1)
if hflip == True:
summ = summ.flip(2)
if rots == 1:
summ = summ.transpose(1,2).flip(1)
if rots == 2:
summ = summ.flip(1).flip(2)
if rots == 3:
summ = summ.transpose(1,2).flip(2)
if norm == True:
s_means = torch.mean(summ)
s_stds = torch.std(summ)
summ -= s_means
summ /= s_stds
if comp_affs == True:
summ = self.GA.get_affs(summ.view(1,1,summ.size(1),summ.size(2)))[:,0,0]
return(summ)
def transform_video(self,video,transform_params, for_eval=False):
hflip,vflip,rots,maxpool_size,x_start,x_stop,y_start,y_stop = transform_params
sec_length = int(video.shape[0] / self.n_secs)
maxpool = nn.MaxPool3d(kernel_size=(maxpool_size,1,1))
if self.predict == False:
sec_starts = np.random.choice(video.shape[0]-sec_length,self.n_secs)
else:
sec_starts = np.arange(int(video.shape[0]/sec_length))*sec_length
corrs = []
for n in range(len(sec_starts)):
video_resized = torch.tensor(video[sec_starts[n]:min(sec_starts[n]+sec_length,video.shape[0]),x_start:x_stop,y_start:y_stop].astype(np.float32),dtype=self.dtype,device=self.device)
video_resized = maxpool(video_resized.reshape(1,1,video_resized.size(0),video_resized.size(1),video_resized.size(2)))[0,0]
if vflip == True:
video_resized = video_resized.flip(1)
if hflip == True:
video_resized = video_resized.flip(2)
if rots == 1:
video_resized = video_resized.transpose(1,2).flip(1)
if rots == 2:
video_resized = video_resized.flip(1).flip(2)
if rots == 3:
video_resized = video_resized.transpose(1,2).flip(2)
corrs.append(self.get_corrs(video_resized))
corrs = torch.stack(corrs)
means = torch.mean(corrs.view(corrs.size(0),corrs.size(1),corrs.size(2)*corrs.size(3)),dim=2).view(corrs.size(0),corrs.size(1),1,1)#,dim=(2,3),keepdim=True)
stds = torch.std(corrs.view(corrs.size(0),corrs.size(1),corrs.size(2)*corrs.size(3)),dim=2).view(corrs.size(0),corrs.size(1),1,1)#,dim=(2,3),keepdim=True)
corrs -= means
corrs /= stds
return(corrs)
def __len__(self):
return(self.data_length)
def __getitem__(self,idx):
video = self.data['videos'][idx]
summary = self.data['summary'][idx]
name = self.data['names'][idx]
summ = torch.tensor(summary.astype(np.float32),dtype=self.dtype,device=self.device)
if self.predict == False:
seg = self.data['segmentations'][idx]
label = self.data['labels'][idx]
# convert to tensors
label = torch.tensor(label.astype(np.float32),dtype=self.dtype,device=self.device)
seg = torch.tensor(seg.astype(np.float32),dtype=self.dtype,device=self.device)
transform_params = self.get_transform_params(seg)
else:
transform_params = self.get_transform_params(summ)
new_summ = self.transform_summlike(summ, transform_params, norm=True, comp_affs=False)
corrs = self.transform_video(video, transform_params)
sample = {'name' : name,
'correlations' : corrs,
'summary' : new_summ
}
if self.predict == False:
affs = self.transform_summlike(seg, transform_params, norm=False, comp_affs=True)
sample['affinities'] = affs
new_label = self.transform_summlike(label, transform_params, norm=False, comp_affs=False)
sample['label'] = new_label
return(sample)
def get_corrs(self,video):
offsets = [[1, 0], [0, 1],[1,1], [2, 0], [0, 2], [2,1], [1,2], [2, 2],
[3,0],[0,3], [3,1], [3,2], [1,3], [2,3], [3,3]]
X = video.size(1)
Y = video.size(2)
corrs = torch.zeros((len(offsets),X,Y),dtype=self.dtype,device=self.device)
u = video
u_ = torch.mean(u,dim=0)
u_u_ = u-u_
u_u_n = torch.sqrt(torch.sum(u_u_**2,dim=0))
for o,off in enumerate(offsets):
v = torch.zeros(video.size(),dtype=self.dtype,device=self.device)
v[:,off[0]:,off[1]:] = video[:,:(video.size(1)-off[0]),:(video.size(2)-off[1])]
v_ = torch.mean(v,dim=0)
v_v_ = v-v_
v_v_n = torch.sqrt(torch.sum(v_v_**2,dim=0))
zaehler = torch.sum(torch.mul(u_u_,v_v_),dim=0)
nenner = torch.mul(u_u_n, v_v_n)
corrs[o] = torch.where(nenner>0.,zaehler.div(nenner),torch.zeros((X,Y),dtype=self.dtype,device=self.device))
return(corrs)