-
Notifications
You must be signed in to change notification settings - Fork 119
/
dataset.py
544 lines (463 loc) · 16.7 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
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
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
# Copyright 2019 Image Analysis Lab, German Center for Neurodegenerative Diseases (DZNE), Bonn
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# IMPORTS
import time
from typing import Optional, Tuple, Dict
import h5py
import numpy as np
import numpy.typing as npt
import torch
import yacs.config
from torch.utils.data import Dataset
from FastSurferCNN.data_loader import data_utils as du
from FastSurferCNN.utils import logging
logger = logging.getLogger(__name__)
# Operator to load imaged for inference
class MultiScaleOrigDataThickSlices(Dataset):
"""
Load MRI-Image and process it to correct format for network inference.
"""
def __init__(
self,
orig_data: npt.NDArray,
orig_zoom: npt.NDArray,
cfg: yacs.config.CfgNode,
transforms: Optional = None
):
"""
Construct object.
Parameters
----------
orig_data : npt.NDArray
Orignal Data.
orig_zoom : npt.NDArray
Original zoomfactors.
cfg : yacs.config.CfgNode
Configuration Node.
transforms : Optional
Transformer for the image. Defaults to None.
"""
assert (
orig_data.max() > 0.8
), f"Multi Dataset - orig fail, max removed {orig_data.max()}"
self.plane = cfg.DATA.PLANE
self.slice_thickness = cfg.MODEL.NUM_CHANNELS // 2
self.base_res = cfg.MODEL.BASE_RES
if self.plane == "sagittal":
orig_data = du.transform_sagittal(orig_data)
self.zoom = orig_zoom[::-1][:2]
logger.info("Loading Sagittal with input voxelsize {}".format(self.zoom))
elif self.plane == "axial":
orig_data = du.transform_axial(orig_data)
self.zoom = orig_zoom[::-1][:2]
logger.info("Loading Axial with input voxelsize {}".format(self.zoom))
else:
self.zoom = orig_zoom[:2]
logger.info("Loading Coronal with input voxelsize {}".format(self.zoom))
# Create thick slices
orig_thick = du.get_thick_slices(orig_data, self.slice_thickness)
orig_thick = np.transpose(orig_thick, (2, 0, 1, 3))
self.images = orig_thick
self.count = self.images.shape[0]
self.transforms = transforms
def _get_scale_factor(self) -> npt.NDArray[float]:
"""
Get scaling factor to match original resolution of input image to final resolution of FastSurfer base network.
Input resolution is taken from voxel size in image header.
ToDO: This needs to be updated based on the plane we are looking at in case we
are dealing with non-isotropic images as inputs.
Returns
-------
npt.NDArray[float]
Scale factor along x and y dimension.
"""
scale = self.base_res / np.asarray(self.zoom)
return scale
def __getitem__(self, index: int) -> Dict:
"""
Return a single image and its scale factor.
Parameters
----------
index : int
Index of image to get.
Returns
-------
dict
Dictionary of image and scale factor.
"""
img = self.images[index]
scale_factor = self._get_scale_factor()
if self.transforms is not None:
img = self.transforms(img)
return {"image": img, "scale_factor": scale_factor}
def __len__(self) -> int:
"""
Return length.
Returns
-------
int
Count.
"""
return self.count
# Operator to load hdf5-file for training
class MultiScaleDataset(Dataset):
"""
Class for loading aseg file with augmentations (transforms).
"""
def __init__(
self,
dataset_path: str,
cfg: yacs.config.CfgNode,
gn_noise: bool = False,
transforms: Optional = None
):
"""
Construct object.
Parameters
----------
dataset_path : str
Path to the dataset.
cfg : yacs.config.CfgNode
Configuration node.
gn_noise : bool
Whether to add gaussian noise (Default value = False).
transforms : Optional
Transformer to apply to the image (Default value = None).
"""
self.max_size = cfg.DATA.PADDED_SIZE
self.base_res = cfg.MODEL.BASE_RES
self.gn_noise = gn_noise
# Load the h5 file and save it to the datase
self.images = []
self.labels = []
self.weights = []
self.subjects = []
self.zooms = []
# Open file in reading mode
start = time.time()
with h5py.File(dataset_path, "r") as hf:
for size in cfg.DATA.SIZES:
try:
logger.info(f"Processing images of size {size}.")
img_dset = list(hf[f"{size}"]["orig_dataset"])
logger.info(
"Processed origs of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
self.images.extend(img_dset)
self.labels.extend(list(hf[f"{size}"]["aseg_dataset"]))
logger.info(
"Processed asegs of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
self.weights.extend(list(hf[f"{size}"]["weight_dataset"]))
self.zooms.extend(list(hf[f"{size}"]["zoom_dataset"]))
logger.info(
"Processed zooms of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
logger.info(
"Processed weights of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
self.subjects.extend(list(hf[f"{size}"]["subject"]))
logger.info(
"Processed subjects of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
logger.info(f"Number of slices for size {size} is {len(img_dset)}")
except KeyError as e:
print(
f"KeyError: Unable to open object (object {size} does not exist)"
)
continue
self.count = len(self.images)
self.transforms = transforms
logger.info(
"Successfully loaded {} data from {} with plane {} in {:.3f} seconds".format(
self.count, dataset_path, cfg.DATA.PLANE, time.time() - start
)
)
def get_subject_names(self):
"""
Get the subject name.
Returns
-------
list
List of subject names.
"""
return self.subjects
def _get_scale_factor(
self,
img_zoom: torch.Tensor,
scale_aug: torch.Tensor
) -> npt.NDArray[float]:
"""
Get scaling factor to match original resolution of input image to final resolution of FastSurfer base network.
Input resolution is taken from voxel size in image header.
ToDO: This needs to be updated based on the plane we are looking at in case we
are dealing with non-isotropic images as inputs.
Parameters
----------
img_zoom : torch.Tensor
Image zoom factor.
scale_aug : torch.Tensor
Scale augmentation factor.
Returns
-------
npt.NDArray[float]
Scale factor along x and y dimension.
"""
if torch.all(scale_aug > 0):
img_zoom *= 1 / scale_aug
scale = self.base_res / img_zoom
if self.gn_noise:
scale += (
torch.randn(1) * 0.1 + 0
) # needs to be changed to torch.tensor stuff
scale = torch.clamp(scale, min=0.1)
return scale
def _pad(
self,
image: npt.NDArray
) -> np.ndarray:
"""
Pad the image with zeros.
Parameters
----------
image : npt.NDArray
Image to pad.
Returns
-------
padded_image
Padded image.
"""
if len(image.shape) == 2:
h, w = image.shape
padded_img = np.zeros((self.max_size, self.max_size), dtype=image.dtype)
else:
h, w, c = image.shape
padded_img = np.zeros((self.max_size, self.max_size, c), dtype=image.dtype)
if self.max_size < h:
sub = h - self.max_size
padded_img = image[0 : h - sub, 0 : w - sub]
else:
padded_img[0:h, 0:w] = image
return padded_img
def unify_imgs(
self,
img: npt.NDArray,
label: npt.NDArray,
weight: npt.NDArray
) -> Tuple[np.ndarray, np.ndarray, np.ndarray]:
"""
Pad img, label and weight.
Parameters
----------
img : npt.NDArray
Image to unify.
label : npt.NDArray
Labels of the image.
weight : npt.NDArray
Weights of the image.
Returns
-------
np.ndarray
Img.
np.ndarray
Label.
np.ndarray
Weight.
"""
img = self._pad(img)
label = self._pad(label)
weight = self._pad(weight)
return img, label, weight
def __getitem__(self, index):
"""
Retrieve processed data at the specified index.
Parameters
----------
index : int
Index to retrieve data for.
Returns
-------
dict
Dictionary containing torch tensors for image, label, weight, and scale factor.
"""
padded_img, padded_label, padded_weight = self.unify_imgs(
self.images[index], self.labels[index], self.weights[index]
)
img = np.expand_dims(padded_img.transpose((2, 0, 1)), axis=3)
label = padded_label[np.newaxis, :, :, np.newaxis]
weight = padded_weight[np.newaxis, :, :, np.newaxis]
import torchio as tio
subject = tio.Subject(
{
"img": tio.ScalarImage(tensor=img),
"label": tio.LabelMap(tensor=label),
"weight": tio.LabelMap(tensor=weight),
}
)
zoom_aug = torch.as_tensor([0.0, 0.0])
if self.transforms is not None:
tx_sample = self.transforms(subject) # this returns data as torch.tensors
img = torch.squeeze(tx_sample["img"].data).float()
label = torch.squeeze(tx_sample["label"].data).byte()
weight = torch.squeeze(tx_sample["weight"].data).float()
# get updated scalefactor, in case of scaling, not ideal - fails if scales is not in dict
rep_tf = tx_sample.get_composed_history()
if rep_tf:
zoom_aug += torch.as_tensor(
rep_tf[0]._get_reproducing_arguments()["scales"]
)[:-1]
# Normalize image and clamp between 0 and 1
img = torch.clamp(img / img.max(), min=0.0, max=1.0)
scale_factor = self._get_scale_factor(
torch.from_numpy(self.zooms[index]), scale_aug=zoom_aug
)
return {
"image": img,
"label": label,
"weight": weight,
"scale_factor": scale_factor,
}
def __len__(self):
"""
Return count.
"""
return self.count
# Operator to load hdf5-file for validation
class MultiScaleDatasetVal(Dataset):
"""
Class for loading aseg file with augmentations (transforms).
"""
def __init__(self, dataset_path, cfg, transforms=None):
self.max_size = cfg.DATA.PADDED_SIZE
self.base_res = cfg.MODEL.BASE_RES
# Load the h5 file and save it to the dataset
self.images = []
self.labels = []
self.weights = []
self.subjects = []
self.zooms = []
# Open file in reading mode
start = time.time()
with h5py.File(dataset_path, "r") as hf:
for size in cfg.DATA.SIZES:
try:
logger.info(f"Processing images of size {size}.")
img_dset = list(hf[f"{size}"]["orig_dataset"])
logger.info(
"Processed origs of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
self.images.extend(img_dset)
self.labels.extend(list(hf[f"{size}"]["aseg_dataset"]))
logger.info(
"Processed asegs of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
self.weights.extend(list(hf[f"{size}"]["weight_dataset"]))
logger.info(
"Processed weights of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
self.zooms.extend(list(hf[f"{size}"]["zoom_dataset"]))
logger.info(
"Processed zooms of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
self.subjects.extend(list(hf[f"{size}"]["subject"]))
logger.info(
"Processed subjects of size {} in {:.3f} seconds".format(
size, time.time() - start
)
)
logger.info(f"Number of slices for size {size} is {len(img_dset)}")
except KeyError as e:
print(
f"KeyError: Unable to open object (object {size} does not exist)"
)
continue
self.count = len(self.images)
self.transforms = transforms
logger.info(
"Successfully loaded {} data from {} with plane {} in {:.3f} seconds".format(
self.count, dataset_path, cfg.DATA.PLANE, time.time() - start
)
)
def get_subject_names(self):
"""
Get subject names.
"""
return self.subjects
def _get_scale_factor(self, img_zoom):
"""
Get scaling factor to match original resolution of input image to final resolution of FastSurfer base network.
Input resolution is taken from voxel size in image header.
ToDO: This needs to be updated based on the plane we are looking at in case we
are dealing with non-isotropic images as inputs.
Parameters
----------
img_zoom : np.ndarray
Voxel sizes of the image.
Returns
-------
np.ndarray : numpy.typing.NDArray[float]
Scale factor along x and y dimension.
"""
scale = self.base_res / img_zoom
return scale
def __getitem__(self, index):
"""
Get item.
"""
img = self.images[index]
label = self.labels[index]
weight = self.weights[index]
scale_factor = self._get_scale_factor(self.zooms[index])
if self.transforms is not None:
tx_sample = self.transforms(
{
"img": img,
"label": label,
"weight": weight,
"scale_factor": scale_factor,
}
)
img = tx_sample["img"]
label = tx_sample["label"]
weight = tx_sample["weight"]
scale_factor = tx_sample["scale_factor"]
return {
"image": img,
"label": label,
"weight": weight,
"scale_factor": scale_factor,
}
def __len__(self):
"""
Get count.
"""
return self.count