-
Notifications
You must be signed in to change notification settings - Fork 3
/
evaluation.py
384 lines (354 loc) · 18.3 KB
/
evaluation.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
from BettiMatching import *
from skimage.transform import rescale
import yaml
import os
import json
from os.path import join
os.environ['KMP_DUPLICATE_LIB_OK']='True'
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
from glob import glob
from argparse import ArgumentParser
import torch
from torch.utils.data import DataLoader
import monai
from monai.data import list_data_collate, decollate_batch
from monai.inferers import sliding_window_inference
from monai.metrics import DiceMetric
from monai.transforms import Activations, AddChanneld, AsDiscrete, Compose, LoadImaged, SaveImage, ScaleIntensityd, EnsureTyped, EnsureType
from monai.networks.nets import UNet
import torchvision
import imageio
from skimage.morphology import skeletonize, skeletonize_3d
import numpy as np
from sklearn.metrics import accuracy_score
from metrics.rand import adapted_rand
from metrics.voi import voi
import pandas as pd
from tqdm import tqdm
class obj:
def __init__(self, dict1):
self.__dict__.update(dict1)
def dict2obj(dict1):
return json.loads(json.dumps(dict1), object_hook=obj)
parser = ArgumentParser()
parser.add_argument('--folder',
default=None,
help='root folder of all the models')
parser.add_argument('--config',
default=None,
help='config file (.yaml) containing the hyper-parameters for training.')
parser.add_argument('--dataconfig',
default=None,
help='data config file (.yaml) containing the dataset specific information.')
parser.add_argument('--cuda_visible_device', nargs='*', type=int, default=[0],
help='list of index where skip conn will be made')
parser.add_argument('--metrics',
default='',
help='metrics to compute (comma separated list of metrics)')
def Dice(prediction, ground_truth):
dice = np.sum(prediction[ground_truth==1])*2.0 / (np.sum(prediction) + np.sum(ground_truth))
return dice
def Accuracy(prediction, ground_truth):
m,n = prediction.shape
acc = np.sum(prediction==ground_truth) / (m*n)
return acc
def normalize(Picture, scale=1, anti_aliasing=True):
Picture = rescale(Picture, scale=scale, anti_aliasing=anti_aliasing)
Picture = Picture / np.max(Picture)
return Picture
def load_model(path, spatial_dims=2, in_channels=1, out_channels=1, channels=(16, 32, 64, 128, 256), strides=(2, 2, 2, 2), num_res_units=2):
device = torch.device("cpu")
model = UNet(
spatial_dims=spatial_dims,
in_channels=in_channels,
out_channels=out_channels,
channels=channels,
strides=strides,
num_res_units=num_res_units,
).to(device)
model.load_state_dict(torch.load(os.getcwd()+'/'+path, map_location=torch.device('cpu'))['model'])
model.eval()
return model
def plot_evaluation(image, models=[], segment=None, data_path='./data/cremi'):
img = imageio.imread(data_path+'/images/image_'+str(image)+'.png')
seg = imageio.imread(data_path+'/labels/label_'+str(image)+'.png')
seg = np.array(seg, dtype=float)
input = torchvision.transforms.functional.to_tensor(np.array(img))
input = input.unsqueeze(0)[:,:,:304,:304]
seg = seg[:304,:304]/np.max(seg)
if segment is not None:
img = img[segment[0][0]:segment[0][1],segment[1][0]:segment[1][1]]
seg = seg[segment[0][0]:segment[0][1],segment[1][0]:segment[1][1]]
outputs = []
outputs_bin = []
for model in models:
output = model(input)
output = torch.squeeze(output)
output = torch.sigmoid(output).detach().numpy()
if segment is not None:
output = output[segment[0][0]:segment[0][1],segment[1][0]:segment[1][1]]
output_bin = ((output>0.5)*1.0)
outputs.append(output)
outputs_bin.append(output_bin)
fig = plt.figure(figsize=(25,25))
plt.tight_layout()
plt.subplots_adjust(left=None, bottom=None, right=None, top=None, wspace=0.01, hspace=0.01)
columns = len(models) + 1
rows = 2
fig.add_subplot(rows, columns, 1)
plt.imshow(img, cmap='gray')
plt.axis('off')
plt.title('Picture')
for i,output in enumerate(outputs):
fig.add_subplot(rows, columns, i+2)
plt.imshow(output, cmap='gray')
plt.axis('off')
plt.title('Output'+str(i))
fig.add_subplot(rows, columns, len(models)+2)
plt.imshow(seg, cmap='gray')
plt.axis('off')
plt.title('Ground Truth')
for i,output_bin in enumerate(outputs_bin):
fig.add_subplot(rows, columns, len(models)+3+i)
plt.imshow(output_bin, cmap='gray')
plt.title('Prediction'+str(i))
plt.axis('off')
plt.show()
metrics = pd.DataFrame(columns=['Acc', 'Dice', 'BNerr', 'BMerr', 'BNerr 0', 'BMerr 0', 'BNerr 1', 'BMerr 1'])
for i,output_bin in enumerate(outputs_bin):
BM = BettiMatching(output_bin, seg, filtration='superlevel')
metrics.loc['Model '+str(i)] = {'Acc': Accuracy(output_bin, seg), 'Dice': Dice(output_bin, seg), 'Betti': BM.Betti_number_error(), 'Betti 0': BM.Betti_number_error(dimensions=[0]), 'Betti 1': BM.Betti_number_error(dimensions=[1]), 'BM': BM.loss(), 'BM 0': BM.loss(dimensions=[0]), 'BM 1': BM.loss(dimensions=[1])}
print(metrics)
return
def compute_metrics(t, relative=False, comparison='union', filtration='superlevel', construction='V'):
BM = BettiMatching(t[0], t[1], relative=relative, comparison=comparison, filtration=filtration, construction=construction)
return BM.loss(dimensions=[0,1]), BM.loss(dimensions=[0]), BM.loss(dimensions=[1]), BM.Betti_number_error(threshold=0.5, dimensions=[0,1]), BM.Betti_number_error(threshold=0.5, dimensions=[0]), BM.Betti_number_error(threshold=0.5, dimensions=[1])
def cl_score(v, s):
"""[this function computes the skeleton volume overlap]
Args:
v ([bool]): [image]
s ([bool]): [skeleton]
Returns:
[float]: [computed skeleton volume intersection]
"""
return np.sum(v*s)/np.sum(s)
def clDice(v_p, v_l):
"""[this function computes the cldice metric]
Args:
v_p ([bool]): [predicted image]
v_l ([bool]): [ground truth image]
Returns:
[float]: [cldice metric]
"""
if len(v_p.shape)==2:
tprec = cl_score(v_p,skeletonize(v_l))
tsens = cl_score(v_l,skeletonize(v_p))
elif len(v_p.shape)==3:
tprec = cl_score(v_p,skeletonize_3d(v_l))
tsens = cl_score(v_l,skeletonize_3d(v_p))
return 2*tprec*tsens/(tprec+tsens)
def evaluation(model, dataconfig, metrics=[], save_path=None, relative=True, comparison='union', filtration='superlevel', pixel_dimension=0):
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device)
data_path = dataconfig.DATA.DATA_PATH
images = sorted(glob(os.path.join(data_path+'images', "*"+dataconfig.DATA.FORMAT)))
segs = sorted(glob(os.path.join(data_path+'labels', "*"+dataconfig.DATA.FORMAT)))
val_files = [{"img": img, "seg": seg} for img, seg in zip(images[-dataconfig.DATA.VAL_SAMPLES:], segs[-dataconfig.DATA.VAL_SAMPLES:])]
val_transforms = Compose(
[
LoadImaged(keys=["img", "seg"]),
AddChanneld(keys=["img", "seg"]),
ScaleIntensityd(keys=["img", "seg"]),
EnsureTyped(keys=["img", "seg"]),
]
)
val_ds = monai.data.Dataset(data=val_files, transform=val_transforms)
val_loader = DataLoader(val_ds, batch_size=8, num_workers=4, collate_fn=list_data_collate)
dice_metric = DiceMetric(include_background=True, reduction="mean", get_not_nans=False)
post_trans = Compose([EnsureType(), Activations(sigmoid=True), AsDiscrete(threshold=0.5)])
#saver = SaveImage(output_dir="./output", output_ext=".png", output_postfix="seg")
with torch.no_grad():
if not os.path.exists(save_path):
metrics_dic = {}
else:
metrics_dic = np.load(save_path, allow_pickle=True).item()
if type(metrics_dic) != dict:
metrics_dic = {}
metrics = []
else:
remove_list = ['dice', 'dice std', 'cldice', 'cldice std', 'accuracy', 'accuracy std',
'betti error', 'betti error std',
'betti_0 error', 'betti_0 error std', 'betti_1 error',
'betti_1 error std', 'ari', 'ari std', 'voi', 'voi std']
[metrics_dic.pop(key, None) for key in remove_list]
losses = []
losses_0 = []
losses_1 = []
betti_errors = []
betti_0_errors = []
betti_1_errors = []
cldices = []
accuracies = []
aris = []
vois = []
vois_ignore_0 = []
for val_data in val_loader:
val_images, val_labels = val_data["img"].to(device), val_data["seg"].to(device)
# define sliding window size and batch size for windows inference
roi_size = tuple(dataconfig.DATA.IMG_SIZE)
sw_batch_size = 16
if dataconfig.DATA.IN_CHANNELS == 1:
val_outputs = sliding_window_inference(val_images, roi_size, sw_batch_size, model)
elif dataconfig.DATA.IN_CHANNELS == 3:
val_outputs = sliding_window_inference(torch.squeeze(val_images).permute(0,3,1,2), roi_size, sw_batch_size, model)
val_outputs = [post_trans(i).cpu() for i in decollate_batch(val_outputs)]
val_labels = decollate_batch(val_labels.cpu())
# compute metric for current iteration
if (len(metrics)==0 or 'Dice' in metrics) and 'Dice' not in metrics_dic.keys():
dice_metric(y_pred=val_outputs, y=val_labels)
for pair in zip(val_outputs,val_labels):
if (len(metrics) == 0 or 'ClDice' in metrics) and 'ClDice' not in metrics_dic.keys():
cldice = clDice(pair[0].numpy(),pair[1].numpy())
cldices.append(cldice)
if (len(metrics) == 0 or 'Betti matching error' in metrics or 'Betti number error' in metrics) and ('Betti matching error' not in metrics_dic.keys() or 'Betti number error' not in metrics_dic.keys()):
loss, loss_0, loss_1, betti_err, betti_0_err, betti_1_err = compute_metrics(pair, relative=relative, comparison=comparison, filtration=filtration, construction='V')
losses.append(loss)
losses_0.append(loss_0)
losses_1.append(loss_1)
betti_errors.append(betti_err)
betti_0_errors.append(betti_0_err)
betti_1_errors.append(betti_1_err)
if (len(metrics) == 0 or 'accuracy' in metrics) and 'Accuracy' not in metrics_dic.keys():
accuracy = accuracy_score(pair[0].numpy().flatten(), pair[1].numpy().flatten())
accuracies.append(accuracy)
if (len(metrics) == 0 or 'ARI' in metrics) and 'ARI' not in metrics_dic.keys():
ari = adapted_rand(np.int32(pair[0].numpy()),np.int32(pair[1].numpy()))
aris.append(ari)
if (len(metrics) == 0 or 'VOI' in metrics) and 'ARI' not in metrics_dic.keys():
voi_score = voi(np.int32(pair[0].numpy()),np.int32(pair[1].numpy()), ignore_groundtruth=[])
vois.append(voi_score)
voi_ignore_0 = voi(np.int32(pair[0].numpy()),np.int32(pair[1].numpy()), ignore_groundtruth=[0])
vois_ignore_0.append(voi_ignore_0)
#for val_output in val_outputs:
# saver(val_output)
# aggregate the final mean dice result
if (len(metrics) == 0 or 'Dice' in metrics) and 'Dice' not in metrics_dic.keys():
dice = dice_metric.aggregate().item()
dice_std = torch.std(dice_metric.get_buffer()).item()
print("Dice:", dice)
print("Dice std", dice_std)
metrics_dic['Dice'] = dice
metrics_dic['Dice std'] = dice_std
if (len(metrics) == 0 or 'ClDice' in metrics) and 'ClDice' not in metrics_dic.keys():
Cldice = np.mean(cldices)
Cldice_std = np.std(cldices)
print("ClDice", Cldice)
print("ClDice std", Cldice_std)
metrics_dic['ClDice'] = Cldice
metrics_dic['ClDice std'] = Cldice_std
if (len(metrics) == 0 or 'Accuracy' in metrics) and 'Accuracy' not in metrics_dic.keys():
Accuracy = np.mean(accuracies)
Accuracy_std = np.std(accuracies)
print("Accuracy", Accuracy)
print("Accuracy std", Accuracy_std)
metrics_dic['Accuracy'] = Accuracy
metrics_dic['Accuracy std'] = Accuracy_std
if (len(metrics) == 0 or 'Betti matching error' in metrics) and 'Betti matching error' not in metrics_dic.keys():
BME = torch.mean(torch.stack(losses))
BME_std = torch.std(torch.stack(losses))
BME_0 = torch.mean(torch.stack(losses_0))
BME_0_std = torch.std(torch.stack(losses_0))
BME_1 = torch.mean(torch.stack(losses_1))
BME_1_std = torch.std(torch.stack(losses_1))
print("Betti matching error:", torch.squeeze(BME).item())
print("Betti matching error std", torch.squeeze(BME_std).item())
print("Betti matching error dim 0", torch.squeeze(BME_0).item())
print("Betti matching error dim 0 std", torch.squeeze(BME_0_std).item())
print("Betti matching error dim 1", torch.squeeze(BME_1).item())
print("Betti matching error dim 1 std", torch.squeeze(BME_1_std).item())
metrics_dic['Betti matching error'] = torch.squeeze(BME).item()
metrics_dic['Betti matching error std'] = torch.squeeze(BME_std).item()
metrics_dic['Betti matching error dim 0'] = torch.squeeze(BME_0).item()
metrics_dic['Betti matching error dim 0 std'] = torch.squeeze(BME_0_std).item()
metrics_dic['Betti matching error dim 1'] = torch.squeeze(BME_1).item()
metrics_dic['Betti matching error dim 1 std'] = torch.squeeze(BME_1_std).item()
if (len(metrics) == 0 or 'Betti number error' in metrics) and 'Betti number error' not in metrics_dic.keys():
Betti_error = np.mean(betti_errors)
Betti_error_std = np.std(betti_errors)
Betti_0_error = np.mean(betti_0_errors)
Betti_0_error_std = np.std(betti_0_errors)
Betti_1_error = np.mean(betti_1_errors)
Betti_1_error_std = np.std(betti_1_errors)
print("Betti number error", Betti_error)
print("Betti number error std", Betti_error_std)
print("Betti number error dim 0", Betti_0_error)
print("Betti number error dim 0 std", Betti_0_error_std)
print("Betti number error dim 1", Betti_1_error)
print("Betti number error dim 1 std", Betti_1_error_std)
metrics_dic['Betti number error'] = Betti_error
metrics_dic['Betti number error std'] = Betti_error_std
metrics_dic['Betti number error dim 0'] = Betti_0_error
metrics_dic['Betti number error dim 0 std'] = Betti_0_error_std
metrics_dic['Betti number error dim 1'] = Betti_1_error
metrics_dic['Betti number error dim 1 std'] = Betti_1_error_std
if (len(metrics) == 0 or 'ARI' in metrics) and 'ARI' not in metrics_dic.keys():
Ari = np.mean(aris)
Ari_std = np.std(aris)
print("ARI", Ari)
print("ARI std", Ari_std)
metrics_dic['ARI'] = Ari
metrics_dic['ARI std'] = Ari_std
if (len(metrics) == 0 or 'VOI' in metrics) and 'VOI' not in metrics_dic.keys():
Voi = np.mean(vois)
Voi_std = np.std(vois)
Voi_ignore_0 = np.mean(vois_ignore_0)
Voi_ignore_0_std = np.std(vois_ignore_0)
print("VOI", Voi)
print("VOI std", Voi_std)
print("VOI ignore 0", Voi_ignore_0)
print("VOI ignore 0 std", Voi_ignore_0_std)
metrics_dic['VOI'] = Voi
metrics_dic['VOI std'] = Voi_std
metrics_dic['VOI_ignore_0'] = Voi_ignore_0
metrics_dic['VOI_ignore_0 std'] = Voi_ignore_0_std
np.save(save_path,metrics_dic)
return
def load_evaluations(folder, key='', models='last'):
assert models in ['best','last']
path_file = os.path.dirname(__file__)+'/models/cremi'
evaluations_df = pd.DataFrame()
for path, subdirs, files in os.walk(folder):
for file in files:
if file.endswith(".npy") and file.startswith(models):
if key in path:
evaluation = np.load(os.path.join(path,file), allow_pickle=True).item()
evaluation_df = pd.DataFrame(evaluation, index=[os.path.relpath(os.path.join(path,file),path_file)])
evaluations_df = pd.concat([evaluations_df,evaluation_df])
return evaluations_df
def main(args):
# Load the dataconfig files
with open(args.config) as f:
print('\n*** Config file')
print(args.dataconfig)
config = yaml.load(f, Loader=yaml.FullLoader)
config = dict2obj(config)
with open(args.dataconfig) as f:
print('\n*** Dataconfig file')
print(args.dataconfig)
dataconfig = yaml.load(f, Loader=yaml.FullLoader)
dataconfig = dict2obj(dataconfig)
files = os.listdir(args.folder)
metrics = [metric for metric in args.metrics.split(',')]
for path, subdirs, files in os.walk(args.folder):
for file in files:
if file.endswith(".pth") and file.startswith('last'):
print(os.path.join(path, file))
model = load_model(os.path.join(path, file), spatial_dims=dataconfig.DATA.DIM, in_channels=dataconfig.DATA.IN_CHANNELS, out_channels=dataconfig.DATA.OUT_CHANNELS, channels=config.MODEL.CHANNELS, strides=config.MODEL.STRIDES, num_res_units=config.MODEL.NUM_RES_UNITS)
save_path = os.path.join(path, file)[:-3]+'npy'
evaluation(model, dataconfig, save_path=save_path, metrics=args.metrics)
print('-----------------------------Evaluation Done-------------------------------------')
if __name__ == "__main__":
args = parser.parse_args()
if args.cuda_visible_device is not None:
os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(map(str, args.cuda_visible_device))
main(args)