-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathgenerate_benchmarks.py
237 lines (218 loc) · 9.81 KB
/
generate_benchmarks.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
226
227
228
229
230
231
232
233
234
235
236
237
import re
import json
import math
files = ["day_sunny","day_overcast","day_rain",
"day_snow","dusk_sunny","dusk_overcast",
"dusk_rain","dusk_snow", "night_sunny", "night_overcast",
"night_rain", "night_snow"]
results = open("results/results_yolo2cat.txt", "w")
for f in files:
f1 = open("result_yolo_" + f + ".json","r")
f2 = open(f + ".json","r")
data1 = json.load(f1)
data2 = json.load(f2)
classes = []
classes.append(100)
classes.append(150)
classes.append(200)
classes.append(250)
classes.append(300)
classes.append(400)
classes.append(500)
classes.append(600)
classes.append(700)
classes.append(800)
classes.append(900)
classes.append(1000)
classes.append(1300)
classes.append(1600)
classes.append(2000)
classes.append(3000)
classes.append(4000)
classes.append(5000)
classes.append(7000)
classes.append(10000)
classes.append(20000)
classes.append(30000)
classes.append(1000000)
# metrics - trueP/ (trueP + falseP) - precision
# - trueP/ (trueP + falseN) - hit rate
# - trueP/ (trueP +falseN + falseP) - accuracy
# for trueP : sum ( x1-x2)^2 + (y1-y2)^2 + (sqrt(w1)-sqrt(w2))^2 + (sqrt(h1)-sqrt(h2))^2
labels = {}
labels2 = {}
sz1 = 0
# print (len(data1))
for entry in data1:
sz1 += len(entry['labels'])
for x in entry['labels']:
if not labels.get(x['name']):
labels[x['name']] = 1
print(x['name'])
# print (sz1)
sz2 = 0
for entry in data2:
for x in entry['labels']:
if x.get('box2d'):
sz2+=1
if not labels2.get(x['category']):
labels2[x['category']] = 1
# print(x['category'])
#print (sz2)
#print("")
dict1 = {}
dict1['car'] = 'car'
dict1['traffic sign'] = 'stop sign'
dict1['traffic light'] = 'traffic light'
dict1['person'] = 'person'
dict1['motor'] = 'motorbike'
dict1['bus'] = 'bus'
dict1['truck'] = 'truck'
dict1['bike'] = 'bicycle'
dict1['train'] = 'train'
dict1['rider'] = 'person'
sz2 = 0
#exit()
IoU = []
for i in range(19):
IoU.append(0.05*(i+1))
print (IoU)
MaP = 0
RaP = 0
for iou in IoU:
trueP = 0
falseP1 = 0
falseP2 = 0
falseN = 0
truepsz = []
falsepsz = []
falsensz = []
for i in range(23):
truepsz.append(0)
falsepsz.append(0)
falsensz.append(0)
error = 0
for entry in data1: # current guess
#print (entry['name'])
for entry2 in data2: #ground truth
if entry2['name'] == entry['name']: #same image
vect = []
ok = []
wgts= []
hgts = []
possibleFalseP = []
for x in entry['labels']:
wgts.append(int(float(x['width'])))
hgts.append(int(float(x['height'])))
vect.append(0)
possibleFalseP.append(0)
if x['name'] == 'car' or x['name'] == 'person':
ok.append(1)
else:
ok.append(0)
for x in entry2['labels']: #iterate trough ground truth labels for current image
real_name = x['category']
# print (x['attributes']['occluded'])
if x.get('box2d') and (x['category'] == 'car' or x['category'] == 'person'):
#print (x['attributes']['occluded'] == False)
#if x['attributes']['occluded'] == False:
sz2+=1
x1r = float(x['box2d']['x1'])
y1r = float(x['box2d']['y1'])
x2r = float(x['box2d']['x2'])
y2r = float(x['box2d']['y2'])
hr = abs(y1r-y2r)
wr = abs(x1r-x2r)
# print (x1r, y1r, x2r, y2r)
procmax = 0
imax = -1
entrygood = {}
i = 0
for x2 in entry['labels']: #search curent label in the guess
if True: #@dict1[real_name] == x2['name']:
xi1 = max(float(x1r), float(x2['x']))
xi2 = min(float(x2r), float(x2['x'])+float(x2['width']))
yi1 = max(float(y1r), float(x2['y']))
yi2 = min(float(y2r), float(x2['y'])+float(x2['height']))
if xi1 <= xi2 and yi1 <=yi2 and (xi2-xi1)*(yi2-yi1)/ ( (x2r-x1r)*(y2r-y1r) + float(x2['width'])*float(x2['height'] ) - (xi2-xi1)*(yi2-yi1) ) > procmax and vect[i] ==0:
if dict1[real_name] == x2['name']:
procmax = (xi2-xi1)*(yi2-yi1)/ ( (x2r-x1r)*(y2r-y1r) + float(x2['width'])*float(x2['height'] ) - (xi2-xi1)*(yi2-yi1))
entrygood = x2
imax = i
else:
possibleFalseP[i] = 1
i = i+1
#print (procmax)
if procmax >= iou:
x1g = float(entrygood['x'])
x2g = float(entrygood['x']) + float(entrygood['width'])
y1g = float(entrygood['y'])
y2g = float(entrygood['y']) + float(entrygood['height'])
hg = float(entrygood['height'])
wg = float(entrygood['width'])
#print (x1g,y1g,hg,wg)
vect[imax] = 1
trueP += 1
#print (trueP)
error +=(x1g-x1r)**2 + (y1g-y1r)**2 + ( math.sqrt(wr) - math.sqrt(wg)) ** 2 + ( math.sqrt(hg) - math.sqrt(hr))** 2
for tt in range(23):
if int(hr*wr)<classes[tt]:
truepsz[tt]+=1
break
else:
falseN += 1
for tt in range(23):
if int(hr*wr)<classes[tt]:
falsensz[tt]+=1
break
for i in range(len(vect)):
if vect[i] == 0 and ok[i] ==1: # false positivie
if possibleFalseP[i] == 1:
falseP1 +=1
else:
falseP2 +=1
for tt in range(23):
if int(hgts[i]*wgts[i])<classes[tt]:
falsepsz[tt]+=1
break
break
falseP = falseP1+ falseP2
MaP += trueP/(trueP+falseP)
RaP += trueP/(trueP + falseN)
results.write("file: " + f + "\n");
results.write("IoU = " + str(iou) + "\n")
totaltp=0
totalfn=0
for i in range(23):
# print (str(classes[i]))
print (truepsz[i])
print(falsensz[i])
totaltp+=truepsz[i]
totalfn+=falsensz[i]
print(totaltp)
print(totalfn)
if truepsz[i]+falsensz[i]>0:
# print ( str(truepsz[i]/(truepsz[i]+falsensz[i])))
stri = "recall for class " + str(classes[i]) + " is " + str(truepsz[i]/(truepsz[i]+falsensz[i])) + "\n"
results.write (stri)
for i in range(23):
if truepsz[i]+falsepsz[i]>0:
stri = "precision for class " + str(classes[i]) + " is " + str(truepsz[i]/(truepsz[i]+falsepsz[i])) + "\n"
results.write (stri)
print("file " + f);
results.write("falseP1 = " + str(falseP1) + "\nfalseP2 = " + str(falseP2) + "\n")
results.write ("trueP: " + str(trueP) + "\n")
results.write ("falseP: "+str(falseP)+ "\n")
results.write ("falseN: "+str(falseN)+ "\n")
if trueP+falseN>0:
results.write ("recall = " + str(trueP/(trueP+falseN))+ "\n")
if trueP+falseP>0:
results.write ("precission = " + str(trueP/(trueP+falseP))+ "\n")
if trueP > 0:
results.write ("average prediction error: "+ str(error/ trueP) + "\n")
results.flush()
MaP = MaP / len(IoU)
RaP = RaP / len(IoU)
results.write("Map = " + str(MaP) + "\n")
results.write("Rap = " + str(RaP) + "\n")
results.close()