-
Notifications
You must be signed in to change notification settings - Fork 5
/
evaluate_crack.py
143 lines (119 loc) · 5.01 KB
/
evaluate_crack.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
import time
import argparse
import codecs
import yaml
from tqdm import tqdm
from newloader import *
from pathlib import Path
from matplotlib import pyplot as plt
from torch.utils.data import DataLoader
from model.TransMUNet import TransMUNet
from utils.utils import get_img_patches, merge_pred_patches
parser = argparse.ArgumentParser()
parser.add_argument('--output', type=str, default='./results.prf')
parser.add_argument('--thresh_step', type=float, default=0.01)
args = parser.parse_args()
def cal_prf_metrics(pred_list, gt_list, thresh_step=0.01):
final_accuracy_all = []
for thresh in np.arange(0.0, 1.0, thresh_step):
# print(thresh)
statistics = []
for pred, gt in zip(pred_list, gt_list):
gt_img = (gt).astype('uint8')
pred_img = (pred > thresh).astype('uint8')
# calculate each image
statistics.append(get_statistics(pred_img, gt_img))
# get tp, fp, fn
tp = np.sum([v[0] for v in statistics])
fp = np.sum([v[1] for v in statistics])
fn = np.sum([v[2] for v in statistics])
# calculate precision
p_acc = 1.0 if tp==0 and fp==0 else tp/(tp+fp)
# calculate recall
r_acc = tp/(tp+fn)
# calculate f-score
final_accuracy_all.append([thresh, p_acc, r_acc, 2*p_acc*r_acc/(p_acc+r_acc)])
return final_accuracy_all
def get_statistics(pred, gt):
"""
return tp, fp, fn
"""
tp = np.sum((pred==1)&(gt==1))
fp = np.sum((pred==1)&(gt==0))
fn = np.sum((pred==0)&(gt==1))
return [tp, fp, fn]
def save_results(input_list, output_path):
with codecs.open(output_path, 'w', encoding='utf-8') as fout:
for ll in input_list:
line = '\t'.join(['%.4f'%v for v in ll])+'\n'
fout.write(line)
def save_sample(img_path, msk, msk_pred, name=''):
img = cv2.imread(img_path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
msk = msk.astype(int)
mskp = msk_pred
_, axs = plt.subplots(1, 3, figsize=(15,5))
axs = axs.ravel()
axs[0].axis('off')
axs[0].imshow(img/255.)
axs[1].axis('off')
axs[1].imshow(msk*255, cmap= 'gray')
axs[2].axis('off')
axs[2].imshow(mskp*255, cmap= 'gray')
plt.savefig(config['save_result'] + name + '.png')
config = yaml.load(open('./config_crack.yml'), Loader=yaml.FullLoader)
number_classes = int(config['number_classes'])
device = 'cuda' if torch.cuda.is_available() else 'cpu'
data_path = config['path_to_testdata']
DIR_IMG = os.path.join(data_path, 'images')
DIR_MASK = os.path.join(data_path, 'masks')
img_names = [path.name for path in Path(DIR_IMG).glob('*.jpg')]
mask_names = [path.name for path in Path(DIR_MASK).glob('*.png')]
test_dataset = Crack_loader(img_dir=DIR_IMG, img_fnames=img_names, mask_dir=DIR_MASK, mask_fnames=mask_names)
test_loader = DataLoader(test_dataset, batch_size = 1, shuffle= False)
print(f'test_dataset:{len(test_dataset)}')
Net = TransMUNet(n_classes = number_classes)
Net = Net.to(device)
Net.load_state_dict(torch.load(config['saved_model'], map_location='cpu')['model_weights'])
pred_list = []
gt_list = []
save_samples = True # if save_samples=False, no samples will be saved.
with torch.no_grad():
print('val_mode')
val_loss = 0
times =0
Net.eval()
for itter, batch in enumerate(tqdm(test_loader)):
img = batch['image'].numpy().squeeze(0)
img_path = batch['img_path'][0]
msk = batch['mask']
patch_totensor = ImgToTensor()
preds = []
start = time.time()
patches, patch_locs = get_img_patches(img)
for i, patch in enumerate(patches):
patch_n = patch_totensor(Image.fromarray(patch)) # torch.Size([3, 256, 256])
X = (patch_n.unsqueeze(0)).to(device, dtype=torch.float) # torch.Size([1, 3, 256, 256])
msk_pred = torch.sigmoid(Net(X)) # torch.Size([1, 1, 256, 256])
mask = msk_pred.cpu().detach().numpy()[0, 0] # (256, 256)
preds.append(mask)
mskp = merge_pred_patches(img, preds, patch_locs) # (H, W)
kernel = np.array(
[
[0, 0, 1, 0, 0],
[0, 1, 1, 1, 0],
[1, 1, 1, 1, 1],
[0, 1, 1, 1, 0],
[0, 0, 1, 0, 0],
], dtype=np.uint8)
mskp = cv2.morphologyEx(mskp, cv2.MORPH_CLOSE, kernel,iterations=1).astype(float)
end = time.time()
times += (end - start)
if itter < 237 and save_samples:
save_sample(img_path, msk.numpy()[0, 0], mskp, name=str(itter+1))
gt_list.append(msk.numpy()[0, 0])
pred_list.append(mskp)
print('Running time of each images: %ss' % (times/len(pred_list)))
final_results = []
final_results = cal_prf_metrics(pred_list, gt_list, args.thresh_step)
save_results(final_results, args.output)