-
Notifications
You must be signed in to change notification settings - Fork 0
/
predict.py
207 lines (157 loc) · 6.22 KB
/
predict.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
import cv2
import time
import torch
import torchvision.utils as vutils
import numpy as np
from PIL import Image
from snet import SNet
def computeIoU(scores, labels):
#scores [4,256,480]
iou = 0
m = scores.shape[0]
x = scores.shape[1]
y = scores.shape[2]
for i in range(m):
#scores = scores.cpu().detach().numpy()
#labels = labels.cpu().detach().numpy()
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==0, scores[i,:,:]==0)
num00 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==0, scores[i,:,:]==1)
num01 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==0, scores[i,:,:]==2)
num02 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==1, scores[i,:,:]==0)
num10 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==1, scores[i,:,:]==1)
num11 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==1, scores[i,:,:]==2)
num12 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==2, scores[i,:,:]==0)
num20 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==2, scores[i,:,:]==1)
num21 = np.count_nonzero(count)
count = np.zeros([x,y],dtype = np.uint8)
count = np.logical_and(labels[i,:,:]==2, scores[i,:,:]==2)
num22 = np.count_nonzero(count)
if num00 == 0:
iou0 = 0
else:
iou0 = num00/(num00+num01+num02+num10+num20)
if num11 ==0:
iou1 =0
else:
iou1 = num11/(num10+num11+num12+num01+num21)
if num22==0:
iou2 = 0
else:
iou2 = num22/(num20+num21+num22+num02+num12)
miou = (iou0 + iou1 + iou2)/3
iou = iou+miou
iou = iou/m
return iou
# def add_Gaussian(flow):
# flow = flow/255
# row,col,ch= np.shape(flow)
# mean = 0
# sigma = 0.5
# gauss = np.random.normal(mean,sigma,(row,col,ch))
# gauss = gauss.reshape(row,col,ch)
# noisy = flow + gauss
# noisy = np.clip(noisy, 0, 1)
# noisy = np.uint8(noisy*255)
# return noisy
lst = [2, 95, 96, 97, 98, 99, 110, 111, 112, 113, 114, 500, 501, 502, 503, 504, 735, 736, 737, 738,
739, 750, 751, 752, 753, 754, 900, 901, 902, 903, 904, 3146, 3861, 4364, 4411, 5382 ]
# lst = [300, 301, 302, 303, 304, 305, 306, 307, 308, 309]
for id in lst:
IMAGE_PATH = './test/scale/original/'+str(id)+'.png'
FLOW_PATH = './test/scale/flow/' +str(id)+'.png'
# FLOW_PATH = 'dresden1.png'
LABEL_PATH = './test/scale/label/'+ str(id)+'.png'
PATH = './models/snet_tmp6final.pt'
device = torch.cuda.current_device()
model = torch.nn.DataParallel(SNet(6)).to(device)
model.load_state_dict(torch.load(PATH))
criterion = torch.nn.CrossEntropyLoss(reduction='mean')
image = Image.open(IMAGE_PATH).convert('RGB')
image = np.asarray(np.array(image),np.float32)
image = image.transpose((2,0,1))
image = torch.tensor(image)
image = image.to(device)
flow = Image.open(FLOW_PATH).convert('RGB')
flow = np.asarray(np.array(flow),np.float32)
#flow = add_Gaussian(flow)
m = np.shape(flow)[0]
n = np.shape(flow)[1]
flow = flow.transpose((2,0,1))#transpose the H*W*C to C*H*W
flow = torch.tensor(flow)
flow = flow.to(device)
label= cv2.imread(LABEL_PATH)
label = cv2.cvtColor(label, cv2.COLOR_BGR2RGB)
new_label = np.empty([m,n],dtype = np.uint8)
for i in range(m):
for j in range(n):
if label[i, j, 0]!= 0:
new_label[i,j] = 2 #object red
if label[i, j, 1] !=0:
new_label[i,j] = 1 # tool green
if label[i, j, 1]==0 and label[i, j, 0]== 0:
new_label[i,j] = 0 #backgrond black
L = torch.tensor(new_label,dtype=torch.long)
L = L.unsqueeze(0)
L = L.to(device)
IF = torch.cat((image,flow), 0)
IF = IF.unsqueeze(0)
image = image.unsqueeze(0)
scores = model(IF)
scores = torch.argmax(scores, dim = 1)
accuracy = (scores == L).float().mean()
# print(scores.size())
scores = scores.cpu().detach().numpy()
L = L.cpu().detach().numpy()
miou = computeIoU(scores, L)
im_viz = np.zeros([m,n,3])
# print(np.shape(im_viz))
for i in range(256):
for j in range(480):
if scores[0,i,j]==0:
im_viz[i,j,0] = 255
im_viz[i,j,1] = 255
im_viz[i,j,2] = 255
if scores[0,i,j] == 2:
im_viz[i,j,0] = 255
im_viz[i,j,1] = 100
im_viz[i,j,2] = 100
if scores[0,i,j] == 1:
im_viz[i,j,0] = 100
im_viz[i,j,1] = 255
im_viz[i,j,2] = 100
# im_viz[:,:,0] = (scores[0,:,:]==0)*240
# im_viz[:,:,0] = (scores[0,:,:]==2)*90
# im_viz[:,:,1] = (scores[0,:,:]==0)*240
# im_viz[:,:,1] = (scores[0,:,:]==1)*90
# im_viz[:,:,2] = (scores[0,:,:]==0)*240
# im_viz = Image.fromarray(np.uint8(im_viz))
# im_viz = im_viz.save('./test/'+str(id)+'.png')
# background = Image.open(IMAGE_PATH)
# overlay = Image.open('./test/'+str(id)+".png")
# background = background.convert("RGBA")
# overlay = overlay.convert("RGBA")
# new_img = Image.blend(background, overlay, 0.4)
# new_img.save("./test/overlay" +str(id)+".png","PNG")
print(id)
print(accuracy)
print(miou)
strtoprint = ''
strtoprint += str(id) +' '
strtoprint += str(accuracy.cpu().numpy())+ ' '
strtoprint += str(miou)
print(strtoprint, file=open('./test/record.txt', 'a'), flush=True)