-
Notifications
You must be signed in to change notification settings - Fork 0
/
tester.py
220 lines (194 loc) · 9.16 KB
/
tester.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
import time
import cv2
import torch
import os
import numpy as np
import torch.backends.cudnn as cudnn
import torch.nn as nn
import torchvision.transforms.functional as TF
from torch.utils.data import DataLoader
from loguru import logger
from tqdm import tqdm
from trainer import Trainer
from utils.helpers import dir_exists, remove_files, double_threshold_iteration
from utils.metrics import AverageMeter, get_metrics, get_metrics, count_connect_component
import ttach as tta
import torch.nn.functional as F
import time
import pickle
import sys
from patch2img import p2i
from patch2img3 import p2i_3
from dataset import vessel_patch_dataset
def load_matching_weights(model, pretrained_weights):
model_dict = model.state_dict()
pretrained_dict = {k: v for k, v in pretrained_weights.items() if k in model_dict and model_dict[k].size() == v.size()}
model_dict.update(pretrained_dict)
model.load_state_dict(model_dict)
class Tester(Trainer):
def __init__(self, model, loss, CFG, checkpoint=None, test_loader=None, dataset_path=None, show=True, mode=None):
# super(Trainer, self).__init__()
self.loss = loss
self.CFG = CFG
self.test_loader = test_loader
self.model = nn.DataParallel(model.cuda())
self.dataset_path = dataset_path
self.show = show
self.mode = mode
self.ck = checkpoint['state_dict']
self.dir = self.CFG['save_dir'] + self.CFG['dataname']["type"] + "/" + self.CFG['model']["type"] + "/"+self.CFG['model']["type"]+"_pre_save_picture/"
self.dir2 = "/home/s1/ZX/job/Vessel/datasets/"+self.CFG['dataname']["type"]+"/test_pro"
load_matching_weights(self.model, self.ck)
dir_exists(self.dir)
remove_files(self.dir)
cudnn.benchmark = True
def adjust_brightness(self, image, beta):
image_float = image.astype(np.float32)
# 调整亮度
image_bright = cv2.convertScaleAbs(image_float, alpha=1, beta=beta)
return image_bright
def patch_img2pre(self,test_patch_loader):
if self.CFG.tta:
self.model = tta.SegmentationTTAWrapper(
self.model, tta.aliases.d4_transform(), merge_mode='mean')
self.model.eval()
tbar = tqdm(test_patch_loader, ncols=150)
patch_L = []
j_L = []
H = []
# start
with torch.no_grad():
for j, img in enumerate(tbar):
img = img.cuda(non_blocking=True)
# multi outputs
# _, _, _, predict = self.model(img) # u2
# _, _, _, _, predict = self.model(img) # u3
# _,_,predict = self.model(img)
# single outputs
predict = self.model(img)
predict = predict[0,0,...]
predict = torch.sigmoid(predict).cpu().numpy()
heatmap = np.uint8(255 * predict)
patch_L.append(predict)
j_L.append(j)
H.append(heatmap)
if self.CFG.heatmap is True:
return patch_L,j_L,H
else:
return patch_L, j_L
def test(self):
pre_L = []
if "DRIVE" in self.CFG['dataname']["type"]:
self.data_len = 20
self.w = 565
self.h = 584
self.overlap_w = 48
self.overlap_h = 48
elif "STARE" in self.CFG['dataname']["type"]:
self.data_len = 10
self.w = 700
self.h = 605
self.overlap_w = 48
self.overlap_h = 48
elif "CHASEDB1" in self.CFG['dataname']["type"]:
self.data_len = 8
self.w = 999
self.h = 960
self.overlap_w = 72
self.overlap_h = 72
elif "RC_SLO" in self.CFG['dataname']["type"]:
self.data_len = 10
self.w = 360
self.h = 320
self.overlap_w = 36
self.overlap_h = 36
elif "IOSTAR" in self.CFG['dataname']["type"]:
self.data_len = 10
self.w = 1024
self.h = 1024
self.overlap_w = 72
self.overlap_h = 72
elif "CHUAC" in self.CFG['dataname']["type"]:
self.data_len = 10
self.w = 512
self.h = 512
self.overlap_w = 56
self.overlap_h = 56
for i in range(self.data_len):
img_patch_data_path = os.path.join(self.dataset_path,f"{self.mode}_pro",str(i))
save_pre_patch_dir = os.path.join(self.dataset_path,f"save_{self.mode}_patch_dir")
img_patch_dataset = vessel_patch_dataset(img_patch_data_path) # data_process
# print(len(img_patch_dataset))
img_patch__loader = DataLoader(img_patch_dataset, 1,shuffle=False, num_workers=16, pin_memory=True) # 132
if self.CFG.heatmap is True:
patch_L,j_L,H = self.patch_img2pre(test_patch_loader=img_patch__loader)
else:
patch_L, j_L = self.patch_img2pre(test_patch_loader=img_patch__loader)
#### save pre patch
p = os.path.join(save_pre_patch_dir,str(i))
print(p)
dir_exists(p)
remove_files(p)
patch_heat_L = []
if self.CFG.heatmap is True:
for patch,j,heatmap_patch in zip(patch_L,j_L,H):
# heat
p2 = os.path.join(p,f"pre_b{j}.png")
p_heatmap = os.path.join(p, f"pre_heatmap{j}.png")
output_shape = (64, 64)
heatmap_patch = torch.from_numpy(heatmap_patch)
heatmap_patch = heatmap_patch.unsqueeze(0).unsqueeze(0).float()
channel_mean = torch.mean(heatmap_patch, dim=1,
keepdim=True) # channel_max,_ = torch.max(feats,dim=1,keepdim=True)
channel_mean = F.interpolate(channel_mean, size=output_shape, mode='bilinear', align_corners=False)
channel_mean = channel_mean.squeeze(0).squeeze(0).cpu().detach().numpy()
channel_mean = (
((channel_mean - np.min(channel_mean)) / (
np.max(channel_mean) - np.min(channel_mean))) * 255).astype(
np.uint8)
heatmap_patch = cv2.applyColorMap(channel_mean, cv2.COLORMAP_JET)
patch_heat_L.append(heatmap_patch)
cv2.imwrite(p_heatmap, np.uint8(heatmap_patch)) # right
# no heat
cv2.imwrite(p2, np.uint8(patch*255)) # right
# 若路径都对，那就是h&w的顺序不对！！！！！！！！！！！！
# pre = p2i(patch_L,w=565,h=584,overlap_w=48,overlap_h=48) ############change
# heat
pre_heat = p2i_3(patch_heat_L,w=self.w,h=self.h,overlap_w=self.overlap_w,overlap_h=self.overlap_h) ############change
pre = p2i(patch_L,w=self.w,h=self.h,overlap_w=self.overlap_w,overlap_h=self.overlap_h) ############change
# pre = p2i_(save_pre_patch_dir,i,w=565,h=584,overlap_w=48,overlap_h=48) ############change
else:
for patch, j in zip(patch_L, j_L):
# heat
p2 = os.path.join(p, f"pre_b{j}.png")
# no heat
cv2.imwrite(p2, np.uint8(patch * 255)) # right
# 若路径都对，那就是h&w的顺序不对！！！！！！！！！！！！
# pre = p2i(patch_L,w=565,h=584,overlap_w=48,overlap_h=48) ############change
# heat
# pre_heat = p2i_3(patch_heat_L,w=self.w,h=self.h,overlap_w=self.overlap_w,overlap_h=self.overlap_h) ############change
pre = p2i(patch_L, w=self.w, h=self.h, overlap_w=self.overlap_w, overlap_h=self.overlap_h) ############change
# pre = p2i_(save_pre_patch_dir,i,w=565,h=584,overlap_w=48,overlap_h=48) ############change
# for vis
if self.show:
# heat
if self.CFG.heatmap is True:
predict_heat = torch.from_numpy(pre_heat.astype(np.float32)).unsqueeze(0).unsqueeze(0)
predict_heat = predict_heat[0,0,...]
predict_heat = predict_heat.cpu().detach().numpy()
cv2.imwrite(
self.dir+f"pre_heat{i}.png", np.uint8(predict_heat))
# no heat
predict = torch.from_numpy(pre.astype(np.float32)).unsqueeze(0).unsqueeze(0)
predict = predict[0,0,...]
predict_pro = predict.cpu().detach().numpy()
predict_b = np.where(predict.cpu().detach().numpy() >= self.CFG.threshold, 1, 0)
cv2.imwrite(
self.dir+f"pre_pro{i}.png", np.uint8(predict_pro*255))
cv2.imwrite(
self.dir+f"pre_b{i}.png", np.uint8(predict_b*255))
# 未二值化 for metrics
with open(file=os.path.join(self.dir, f'pre_{i}.pkl'), mode='wb') as file:
print(file)
pickle.dump(pre, file)
print(f'save pre_{i}.pkl')