-
Notifications
You must be signed in to change notification settings - Fork 3
/
test_without_TTA_mask_keypoints_link.py
225 lines (210 loc) · 10.2 KB
/
test_without_TTA_mask_keypoints_link.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
import torch
import torch.nn as nn
import torch.utils.data as data
from torch.autograd import Variable as V
import cv2
import os
import numpy as np
import matplotlib.pyplot as plt
import pickle
import scipy.io
from time import time
'''
from networks.unet import Unet
from networks.dunet import Dunet
from networks.dinknet import LinkNet34, DinkNet34, DinkNet50, DinkNet101, DinkNet34_less_pool
'''
from networks.dinknet import DinkNet34_WithBranch
# BATCHSIZE_PER_CARD = 4
class TTAFrame():
def __init__(self, net):
self.net = net().cuda()
self.net = torch.nn.DataParallel(self.net, device_ids=range(torch.cuda.device_count()))
def test_one_img_without_TTAFrame(self, path, evalmode=True):
if evalmode:
self.net.eval()
img = cv2.imread(path)
img = np.expand_dims(np.array(img),axis=0)
img = img.transpose(0,3,1,2)
img = np.array(img, np.float32)/255.0 * 3.2 -1.6
img = V(torch.Tensor(img).cuda())
mask, prob, posi, link = self.net.forward(img)
mask = mask.squeeze().cpu().data.numpy()
prob = prob.squeeze(0).cpu().data.numpy()
posi = posi.squeeze().cpu().data.numpy()
link = link.squeeze().cpu().data.numpy()
return mask, prob, posi, link
def load(self, path):
self.net.load_state_dict(torch.load(path))
source = 'dataset/valid/sat/'
val = os.listdir(source)
solver = TTAFrame(DinkNet34_WithBranch)
solver.load('weights/log04_dink34.th')
tic = time()
target = 'submits/log04_dink34_without_TTA/'
os.mkdir(target)
os.mkdir(target+'mask/')
os.mkdir(target+'mat/')
os.mkdir(target+'prob_posi_link/')
os.mkdir(target+'merge/')
for img_id,name in enumerate(val):
if img_id % 10 == 0:
print(img_id/10, ' ','%.2f'%(time()-tic))
'''
if name != '794635_sat.jpg':
continue
'''
mask, prob, posi, link = solver.test_one_img_without_TTAFrame(source + name)
mask[mask > 0.5] = 255
mask[mask <= 0.5] = 0
prob[prob > 0.5] = 1
prob[prob <= 0.5] = 0
link[link > 0.5] = 1
link[link <= 0.5] = 0
posi_final = np.zeros((2,64,64), np.uint64)
for i in range(0, 64):
for j in range(0, 64):
if prob[0,i,j] == 1:
posi_final[0,i,j] = int(posi[0,i,j]*15+0.5)+i*16
posi_final[1,i,j] = int(posi[1,i,j]*15+0.5)+j*16
# 修正邻接矩阵
if link[0,i,j]==1:
if i-1<0 or prob[0,i-1,j]!=1:
link[0,i,j]=0
if link[1,i,j]==1:
if i-1<0 or j+1>=64 or prob[0,i-1,j+1]!=1:
link[1,i,j]=0
if link[2,i,j]==1:
if j+1>=64 or prob[0,i,j+1]!=1:
link[2,i,j]=0
if link[3,i,j]==1:
if i+1>=64 or j+1>=64 or prob[0,i+1,j+1]!=1:
link[3,i,j]=0
if link[4,i,j]==1:
if i+1>=64 or prob[0,i+1,j]!=1:
link[4,i,j]=0
if link[5,i,j]==1:
if i+1>=64 or j-1<0 or prob[0,i+1,j-1]!=1:
link[5,i,j]=0
if link[6,i,j]==1:
if j-1<0 or prob[0,i,j-1]!=1:
link[6,i,j]=0
if link[7,i,j]==1:
if i-1<0 or j-1<0 or prob[0,i-1,j-1]!=1:
link[7,i,j]=0
# 连接斜向邻接的anchor
if link[1,i,j] == 0 and i-1>=0 and j+1<64:
if prob[0,i-1,j+1] == 1 and prob[0,i-1,j] == 0 and prob[0,i,j+1] == 0:
link[1,i,j] = 1
if link[3,i,j] == 0 and i+1<64 and j+1<64:
if prob[0,i+1,j+1] == 1 and prob[0,i,j+1] == 0 and prob[0,i+1,j] == 0:
link[3,i,j] = 1
if link[5,i,j] == 0 and i+1<64 and j-1>=0:
if prob[0,i+1,j-1] == 1 and prob[0,i+1,j] == 0 and prob[0,i,j-1] == 0:
link[5,i,j] = 1
if link[7,i,j] == 0 and i-1>=0 and j-1>=0:
if prob[0,i-1,j-1] == 1 and prob[0,i,j-1] == 0 and prob[0,i-1,j] == 0:
link[7,i,j] = 1
else:
posi_final[0,i,j] = -1
posi_final[1,i,j] = -1
for k in range(0,8):
link[k,i,j] == -1
# 存在2*2方块互相连接则去掉最长边 (RemoveCircle_Update操作)
posi_cal = posi_final.astype(np.int64)
for i in range(0, 63):
for j in range(0, 63):
if prob[0,i,j] == 1 and prob[0,i+1,j] == 1 and prob[0,i,j+1] == 1 and prob[0,i+1,j+1] == 1 and (link[4,i,j] == 1 or link[0,i+1,j] == 1) and (link[2,i+1,j] == 1 or link[6,i+1,j+1] == 1) and (link[0,i+1,j+1] == 1 or link[4,i,j+1] == 1) and (link[6,i,j+1] == 1 or link[2,i,j] == 1):
# i,j --> i+1,j
a = (posi_cal[0,i,j]-posi_cal[0,i+1,j])*(posi_cal[0,i,j]-posi_cal[0,i+1,j]) + (posi_cal[1,i,j]-posi_cal[1,i+1,j])*(posi_cal[1,i,j]-posi_cal[1,i+1,j])
# i+1,j --> i+1,j+1
b = (posi_cal[0,i+1,j]-posi_cal[0,i+1,j+1])*(posi_cal[0,i+1,j]-posi_cal[0,i+1,j+1]) + (posi_cal[1,i+1,j]-posi_cal[1,i+1,j+1])*(posi_cal[1,i+1,j]-posi_cal[1,i+1,j+1])
# i,j --> i,j+1
c = (posi_cal[0,i,j]-posi_cal[0,i,j+1])*(posi_cal[0,i,j]-posi_cal[0,i,j+1]) + (posi_cal[1,i,j]-posi_cal[1,i,j+1])*(posi_cal[1,i,j]-posi_cal[1,i,j+1])
# i,j+1 --> i+1,j+1
d = (posi_cal[0,i,j+1]-posi_cal[0,i+1,j+1])*(posi_cal[0,i,j+1]-posi_cal[0,i+1,j+1]) + (posi_cal[1,i,j+1]-posi_cal[1,i+1,j+1])*(posi_cal[1,i,j+1]-posi_cal[1,i+1,j+1])
if a>=b and a>=c and a>=d:
link[4,i,j] = 0
link[0,i+1,j] = 0
link[2,i+1,j] = 1
link[6,i+1,j+1] = 1
link[2,i,j] = 1
link[6,i,j+1] = 1
link[4,i,j+1] = 1
link[0,i+1,j+1] = 1
if b>=a and b>=c and b>=d:
link[4,i,j] = 1
link[0,i+1,j] = 1
link[2,i+1,j] = 0
link[6,i+1,j+1] = 0
link[2,i,j] = 1
link[6,i,j+1] = 1
link[4,i,j+1] = 1
link[0,i+1,j+1] = 1
if c>=a and c>=b and c>=d:
link[4,i,j] = 1
link[0,i+1,j] = 1
link[2,i+1,j] = 1
link[6,i+1,j+1] = 1
link[2,i,j] = 0
link[6,i,j+1] = 0
link[4,i,j+1] = 1
link[0,i+1,j+1] = 1
if d>=a and d>=b and d>=c:
link[4,i,j] = 1
link[0,i+1,j] = 1
link[2,i+1,j] = 1
link[6,i+1,j+1] = 1
link[2,i,j] = 1
link[6,i,j+1] = 1
link[4,i,j+1] = 0
link[0,i+1,j+1] = 0
mask = np.concatenate([mask[:,:,None],mask[:,:,None],mask[:,:,None]],axis=2)
cv2.imwrite(target+'mask/'+name[:-7]+'mask.png',mask.astype(np.uint8))
mat_savepath = target+'mat/' + name[:-7] + 'mask.mat'
scipy.io.savemat(mat_savepath, mdict={'if_key_points': prob, 'all_key_points_position': posi_final, 'anchor_link': link})
new_img = np.zeros((1024,1024,3), np.uint8)
for i in range(0, 64):
for j in range(0, 64):
for m in range(16*i,16*i+16):
for n in range(16*j,16*j+16):
new_img[m][n] = [255,255,255]
if prob[0,i,j] == 0:
new_img[m][n] = [0,255,255]
for i in range(0, 64):
for j in range(0, 64):
if prob[0,i,j] == 1:
if link[0,i,j]==1 and i-1>=0: # i,j ---> i-1,j
if prob[0,i-1,j]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i-1,j],posi_final[0,i-1,j]), (0,255,0), 1)
if link[1,i,j]==1 and i-1>=0 and j+1<64: # i,j ---> i-1,j+1
if prob[0,i-1,j+1]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i-1,j+1],posi_final[0,i-1,j+1]), (0,255,0), 1)
if link[2,i,j]==1 and j+1<64: # i,j ---> i,j+1
if prob[0,i,j+1]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i,j+1],posi_final[0,i,j+1]), (0,255,0), 1)
if link[3,i,j]==1 and i+1<64 and j+1<64: # i,j ---> i+1,j+1
if prob[0,i+1,j+1]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i+1,j+1],posi_final[0,i+1,j+1]), (0,255,0), 1)
if link[4,i,j]==1 and i+1<64: # i,j ---> i+1,j
if prob[0,i+1,j]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i+1,j],posi_final[0,i+1,j]), (0,255,0), 1)
if link[5,i,j]==1 and i+1<64 and j-1>=0: # i,j ---> i+1,j-1
if prob[0,i+1,j-1]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i+1,j-1],posi_final[0,i+1,j-1]), (0,255,0), 1)
if link[6,i,j]==1 and j-1>=0: # i,j ---> i,j-1
if prob[0,i,j-1]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i,j-1],posi_final[0,i,j-1]), (0,255,0), 1)
if link[7,i,j]==1 and i-1>=0 and j-1>=0: # i,j ---> i-1,j-1
if prob[0,i-1,j-1]==1:
cv2.line(new_img, (posi_final[1,i,j],posi_final[0,i,j]), (posi_final[1,i-1,j-1],posi_final[0,i-1,j-1]), (0,255,0), 1)
for i in range(0, 64):
for j in range(0, 64):
for m in range(16*i,16*i+16):
for n in range(16*j,16*j+16):
if (prob[0,i,j] == 1) and (posi_final[0,i,j]==m) and (posi_final[1,i,j]==n):
new_img[m][n] = [0,0,255]
cv2.imwrite(target+'prob_posi_link/'+name[:-7]+'prob_posi_link.png', new_img)
sat = cv2.imread(source + name)
sat_merge = cv2.addWeighted(sat, 0.8, new_img, 0.2, 0)
cv2.imwrite(target+'merge/'+name[:-7]+'merge.png', sat_merge)