Repository navigation
Expand file tree
/
Copy patheval.py
More file actions
202 lines (172 loc) · 7.41 KB
/
Copy patheval.py
File metadata and controls
202 lines (172 loc) · 7.41 KB
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
import argparse
import pathlib
import torch
import torch.nn.functional as F
import numpy as np
from PIL import Image
from tqdm import tqdm
import albumentations as A
from albumentations.pytorch.transforms import ToTensorV2
from TransInvNet.model.model import TransInvNet, CONFIGS
class Metrics:
def __init__(self, pred, mask):
self.h, self.w = mask.shape[0], mask.shape[1]
self.origin_pred, self.origin_mask = pred, mask
self.pred, self.mask = pred.flatten(), mask.flatten()
self.intersection = np.sum(self.pred * self.mask)
self.mask_sum = np.sum(np.abs(self.pred)) + np.sum(np.abs(self.mask))
self.union = self.mask_sum - self.intersection
self.abs_error = np.abs(self.pred - self.mask)
def calculate_iou(self):
return (self.intersection + 1e-20) / (self.union + 1e-20)
def calculate_mae(self):
return np.mean(self.abs_error)
def calculate_dice(self):
return (self.intersection * 2 + 1e-20) / (self.mask_sum + 1e-20)
def _S_object(self):
fg = np.where(self.mask == 0, np.zeros_like(self.pred), self.pred)
bg = np.where(self.mask == 1, np.zeros_like(self.pred), 1 - self.pred)
o_fg = self._object(fg, self.mask)
o_bg = self._object(bg, 1 - self.mask)
u = np.mean(self.mask)
Q = u * o_fg + (1 - u) * o_bg
return Q
def _object(self, pred, mask):
temp = pred[mask == 1]
x = np.mean(temp)
sigma_x = np.std(temp)
score = 2.0 * x / (x * x + 1.0 + sigma_x + 1e-20)
return score
def _S_region(self):
X, Y = self._centroid()
gt1, gt2, gt3, gt4, w1, w2, w3, w4 = self._divideGT(X, Y)
p1, p2, p3, p4 = self._dividePrediction(X, Y)
Q1 = self._ssim(p1, gt1)
Q2 = self._ssim(p2, gt2)
Q3 = self._ssim(p3, gt3)
Q4 = self._ssim(p4, gt4)
Q = w1 * Q1 + w2 * Q2 + w3 * Q3 + w4 * Q4
return Q
def _centroid(self):
if self.mask.sum() == 0:
X = np.eye(1) * round(self.w / 2)
Y = np.eye(1) * round(self.h / 2)
else:
total = self.mask.sum()
i = np.arange(0, self.w)
j = np.arange(0, self.h)
X = round((self.origin_mask.sum(axis=0) * i).sum() / total)
Y = round((self.origin_mask.sum(axis=1) * j).sum() / total)
return X, Y
def _divideGT(self, X, Y):
area = self.h * self.w
LT = self.origin_mask[:Y, :X]
RT = self.origin_mask[:Y, X:self.w]
LB = self.origin_mask[Y:self.h, :X]
RB = self.origin_mask[Y:self.h, X:self.w]
w1 = (X * Y) / area
w2 = (self.w - X) * Y / area
w3 = X * (self.h - Y) / area
w4 = 1 - w1 - w2 - w3
return LT, RT, LB, RB, w1, w2, w3, w4
def _dividePrediction(self, X, Y):
LT = self.origin_pred[:Y, :X]
RT = self.origin_pred[:Y, X:self.w]
LB = self.origin_pred[Y:self.h, :X]
RB = self.origin_pred[Y:self.h, X:self.w]
return LT, RT, LB, RB
def _ssim(self, pred, mask):
N = self.h * self.w
x = pred.mean()
y = mask.mean()
sigma_x2 = ((pred - x) * (pred - x)).sum() / (N - 1 + 1e-20)
sigma_y2 = ((mask - y) * (mask - y)).sum() / (N - 1 + 1e-20)
sigma_xy = ((pred - x) * (mask - y)).sum() / (N - 1 + 1e-20)
alpha = 4 * x * y * sigma_xy
beta = (x * x + y * y) * (sigma_x2 + sigma_y2)
if alpha != 0:
Q = alpha / (beta + 1e-20)
elif alpha == 0 and beta == 0:
Q = 1.0
else:
Q = 0
return Q
def calculate_s_measure(self):
mask_mean = np.mean(self.mask)
pred_mean = np.mean(self.pred)
if mask_mean == 0:
Q = 1 - pred_mean
elif mask_mean == 1:
Q = pred_mean
else:
Q = np.maximum(0.5 * self._S_object() + (1 - 0.5) * self._S_region(), 0)
return Q
def _eval_e(self, num=255):
score = np.zeros(num)
for i in range(num):
fm = self.pred - np.mean(self.pred)
gt = self.mask - np.mean(self.mask)
align_matrix = 2 * gt * fm / (gt * gt + fm * fm + 1e-20)
enhanced = ((align_matrix + 1) * (align_matrix + 1)) / 4
score[i] = np.sum(enhanced) / (len(self.mask) - 1 + 1e-20)
return np.max(score)
def calculate_e_measure(self):
max_e = self._eval_e(255)
return max_e
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--img_size', type=int,
default=352, help='training dataset size')
parser.add_argument('--weight_path', type=str,
default='outputs/exp05240731/train/TransInvNet-best.pth', help='path to the trained weight')
parser.add_argument('--test_path', type=str,
default='datasets/polyp-dataset/'
'TestDataset/CVC-ColonDB', help='path to test dataset')
opt = parser.parse_args()
cfg = CONFIGS['ViT-B_16']
model = TransInvNet(cfg, opt.img_size, vis=True).cuda()
model.load_state_dict(torch.load(opt.weight_path))
model.eval()
test_images_path = [i for i in (pathlib.Path(opt.test_path) / 'images').iterdir()]
test_masks_path = [i for i in (pathlib.Path(opt.test_path) / 'masks').iterdir()]
trans = A.Compose([
A.Normalize(mean=[0.497, 0.302, 0.216],
std=[0.320, 0.217, 0.173]),
ToTensorV2(),
])
Miou, Amae, Mdice, Smeasure, Emeasure = [], [], [], [], []
with torch.no_grad():
tbar = tqdm(zip(test_images_path, test_masks_path), desc='\r', total=len(test_images_path))
for img, gt in tbar:
img = Image.open(img).convert('RGB').resize((opt.img_size, opt.img_size))
img = np.array(img)
img = trans(image=img)['image'][None]
img = img.cuda()
gt = np.asarray(Image.open(gt).convert('1')).astype(np.float)
pred = model(img)
pred = F.interpolate(pred, size=gt.shape, mode='bilinear', align_corners=True)
result = pred.sigmoid().cpu().numpy().squeeze()
result = (result - result.min() + 1e-20) / (result.max() - result.min() + 1e-20)
metrics = Metrics(result, gt)
iou = metrics.calculate_iou()
mae = metrics.calculate_mae()
dice = metrics.calculate_dice()
s_measure = metrics.calculate_s_measure()
e_measure = metrics.calculate_e_measure()
Miou.append(iou)
Amae.append(mae)
Mdice.append(dice)
Smeasure.append(s_measure)
Emeasure.append(e_measure)
tbar.set_description('mIOU: {:.4f}, Average MAE: {:.4f}, Mean DICE: {:.4f}, '
'S-Measure {:.4f}, MAX E-Measure {:.4f}'
.format(np.mean(Miou), np.mean(Amae), np.mean(Mdice), np.mean(Smeasure),
np.mean(Emeasure)))
'''
Dataset | mIOU | Average MAE | Mean DICE | S-Measure | Max E-Measure |
Kvasi-SEG | 0.8536 | 0.0269 | 0.9100 | 0.9165 | 0.9527 |
CVC-612 | 0.8363 | 0.0149 | 0.8905 | 0.9223 | 0.9551 |
ETIS | 0.6036 | 0.0166 | 0.6840 | 0.8133 | 0.8576 |
Endoscene | 0.8273 | 0.0059 | 0.8995 | 0.9335 | 0.9734 |
CVC-Colon | 0.6967 | 0.0295 | 0.7831 | 0.8541 | 0.9007 |
'''