-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathconfIoU.py
559 lines (483 loc) · 21.7 KB
/
confIoU.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
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
#!/usr/bin/env python
#import sklearn.naive_bayes#
from multiprocessing import Pool
import os.path
import numpy as np #
from commands import getoutput as go
import sys
import time
import numpy.matlib #
from collections import defaultdict
import cPickle
import scipy.stats#
import cv2#
import matplotlib.pyplot as plt#
def myread(fname):
txt=open(fname).read().strip()
if txt[:3]=='\xef\xbb\xbf':
txt=txt[3:]
return txt
def download(url,fname):
cmd='wget -c '+url+' -O '+fname
go(cmd)
def get2PointIU(gtMat,resMat):
gtMat=gtMat.copy()
resMat=resMat.copy()
maxProposalsIoU=int(switches['maxProposalsIoU'])
if maxProposalsIoU>0:
resMat=resMat[:maxProposalsIoU,:]
#matSz=(gtMat.shape[0],resMat.shape[0])
gtLeft=numpy.matlib.repmat(gtMat[:,0],resMat.shape[0],1)
gtTop=numpy.matlib.repmat(gtMat[:,1],resMat.shape[0],1)
gtRight=numpy.matlib.repmat((gtMat[:,0]+gtMat[:,2])-1,resMat.shape[0],1)
gtBottom=numpy.matlib.repmat((gtMat[:,1]+gtMat[:,3])-1,resMat.shape[0],1)
gtWidth=numpy.matlib.repmat(gtMat[:,2],resMat.shape[0],1)
gtHeight=numpy.matlib.repmat(gtMat[:,3],resMat.shape[0],1)
resLeft=numpy.matlib.repmat(resMat[:,0],gtMat.shape[0],1).T
resTop=numpy.matlib.repmat(resMat[:,1],gtMat.shape[0],1).T
resRight=numpy.matlib.repmat((resMat[:,0]+resMat[:,2])-1,gtMat.shape[0],1).T
resBottom=numpy.matlib.repmat((resMat[:,1]+resMat[:,3])-1,gtMat.shape[0],1).T
resWidth=numpy.matlib.repmat(resMat[:,2],gtMat.shape[0],1).T
resHeight=numpy.matlib.repmat(resMat[:,3],gtMat.shape[0],1).T
intL=np.max([resLeft,gtLeft],axis=0)
intT=np.max([resTop,gtTop],axis=0)
intR=np.min([resRight,gtRight],axis=0)
intB=np.min([resBottom,gtBottom],axis=0)
intW=(intR-intL)+1
intW[intW<0]=0
intH=(intB-intT)+1
intH[intH<0]=0
I=intH*intW
U=resWidth*resHeight+gtWidth*gtHeight-I
minsurf=np.minimum(resHeight*resWidth,gtWidth*gtHeight)
IoU=I/(U+.0000000001)
resTLRB=resMat[:,:]
resTLRB[:,[2,3]]+=resTLRB[:,[0,1]]-1
gtTLRB=gtMat[:,:]
gtTLRB[:,[2,3]]+=gtTLRB[:,[0,1]]-1
return (IoU,I,U,minsurf)
#filename conversions
def createRequiredDirs(filenameList,fromFilesDir):
filesDirList=set(['/'.join(f.split('/')[:-1]) for f in filenameList])
if fromFilesDir[0]=='+':
for fd in filesDirList:
addP=fromFilesDir[1:].split('/')
addP[-1]=addP[-1]+fd.split('/')[-1]
go('mkdir -p '+fd+'/'+'/'.join(addP))
else:
for fd in filesDirList:
go('mkdir -p '+fd+'/'+fromFilesDir)
def getInputFromConf(confFname):
pathList=confFname.split('/')
pathList[-2]='input'
return '/'.join(pathList)[:-3]+'jpg'
def getThresholdFromHm(hmFname,outDir):
pathList=hmFname.split('/')
pathList[-2]=outDir+pathList[-2]
return '/'.join(pathList)[:-3]+'csv'
def getProposalFromConf(hmFname):
pathList=hmFname.split('/')
pathList[-2]=switches['propDir']
return ('/'.join(pathList)) [:-3]+'csv'
def getConfFromHm(hmFname):
pathList=hmFname.split('/')
pathList[-2]='conf_'+pathList[-2]
return '/'.join(pathList)[:-3]+'csv'
def getIouFromConf(confFname):
pathList=confFname.split('/')
pathList[-2]='iou_'+pathList[-2]
return '/'.join(pathList)[:-3]+'png'
def getGtFromConf(imageFname):
pathList=imageFname.split('/')
pathList[-2]='gt'
return '/'.join(pathList)[:-3]+'txt'
def getProposalFromImage(imageFname,thr=0,hm=''):
if thr==0:
pathList=imageFname.split('/')
pathList[-2]=switches['propDir']
return '/'.join(pathList)[:-3]+'csv'
else:
pathList=imageFname.split('/')
pathList[-2]='prop%d%s'%(int(thr*100),hm)
return '/'.join(pathList)[:-3]+'csv'
def getProposalFromHeatmap(heatmapFname):
if heatmapFname[-3:]=='png':
heatmapFname=heatmapFname[:-3]+'csv'
pathList=heatmapFname.split('/')
pathList[-2]=switches['propDir']
return '/'.join(pathList)
def getConfidenseFromHeatmap(heatmapFname):
if heatmapFname[-3:]=='png':
heatmapFname=heatmapFname[:-3]+'csv'
pathList=heatmapFname.split('/')
pathList[-2]='conf_'+pathList[-2]
return '/'.join(pathList)
getConfidenseFromProposal=getConfidenseFromHeatmap
#str 2 numpy
def arrayToCsvStr(arr):
if len(arr.shape)!=2:
raise Exception("only 2D arrays")
resLines=[list(row) for row in list(arr)]
return '\n'.join([','.join([str(col) for col in row]) for row in resLines])
def csvStr2Array(csvStr):
return np.array([[float(c) for c in l.split(',')] for l in csvStr.split('\n') if len(l)>0])
def fname2Array(fname):
if fname[-3:]=='png':
return cv2.imread(fname,cv2.IMREAD_GRAYSCALE)/255.0
else: #assuming csv
if os.path.isfile(fname) and os.path.getsize(fname)>0:
res= np.genfromtxt(fname, delimiter=',')
return res
else:
return np.empty([0,5])
def array2csvFname(arr,csvFname):
np.savetxt(csvFname,arr, '%9.5f',delimiter=',')
def array2pngFname(arr,pngFname):
cv2.imwrite(pngFname,(arr*255).astype('uint8'),[cv2.IMWRITE_PNG_COMPRESSION ,0])
def loadTxtGtFile(fname):
txt=myread(fname)
lines=[l.strip().split(',') for l in txt.split('\n') if len(l.strip())>0]
print '{%s}'%lines
if len(lines)==0:
return np.zeros([0,4]),np.zeros([0],dtype=object)
# if len(lines)==0:
# return (np.array([5,5,3,3],dtype='float'),['###'])
# if lines[0][0][:3]=='\xef\xbb\xbf':
# lines[0][0]=lines[0][0][3:]
for line in lines:
if len(line) < 4:
print 'BAD LINE:',str(line),'\nFile:',fname
raise Exception()
lines.remove(line)
elif line == ['|']:
lines.remove(line)
for line in lines:
if len(line) < 4:
lines.remove(line)
elif line == ['|']:
lines.remove(line)
if len(lines[0])>8:#4 points
rects=np.empty([len(lines),4],dtype='float')
tmpArr=np.array([[int(c) for c in line[:8]] for line in lines])
left=tmpArr[:,[0,2,4,6]].min(axis=1)
right=tmpArr[:,[0,2,4,6]].max(axis=1)
top=tmpArr[:,[1,3,5,7]].min(axis=1)
bottom=tmpArr[:,[1,3,5,7]].max(axis=1)
rects[:,0]=left
rects[:,1]=top
rects[:,2]=1+right-left
rects[:,3]=1+bottom-top
trans=[','.join(line[8:]) for line in lines]
else:#ltwh
rects=np.array([[int(c) for c in line[:4]] for line in lines],dtype='float')
trans=[','.join(line[4:]) for line in lines]
return (rects,trans)
def loadVggTranscription(fname):
txt=myread(fname)
lines=[l.split(',') for l in txt.split('\n')]
LTWHConf=np.empty([len(lines),5])
transcriptions=np.empty([LTWHConf.shape[0]],dtype='object')
transcriptions[:]=[l[-1] for l in lines]
if len(lines) > 1:
LTWHConf[:,:]=np.array([[float(c) for c in l[:-1]] for l in lines])
return LTWHConf,transcriptions
def getDontCare(transcriptions,dictionary=[]):
dictionary=set(dictionary)
if len(dictionary)==0:
return np.array([tr!='###' for tr in transcriptions],dtype='bool')
else:
return np.array([(tr in dictionary) for tr in transcriptions],dtype='bool')
def getNmsIdx(LTWHConf):
#sort rectangles from large to small
#LTWHConf=LTWHConf[np.argsort(LTWHConf[:,4].prod(axis=1))[::-1],:]
(IoU,I,U,minsurf)=get2PointIU(LTWHConf[:,:4],LTWHConf[:,:4])
iouThr=eval(switches['iouThr'])
return np.where((np.triu(IoU,1)>iouThr).sum(axis=0)<1)[0]
#return LTWHConf[(np.triu(IoU,1)>iouThr).sum(axis=0)<1,:]
#algorithm
def getConfidenceForAll(hm,prop):
prop=prop.astype('int32')
ihm=np.zeros([hm.shape[0]+1,hm.shape[1]+1])
ihm[1:,1:]=hm.cumsum(axis=0).cumsum(axis=1) #integral image
confidenceDict=defaultdict(list)
for rectId in range(prop.shape[0]):
rect=tuple(prop[rectId,:4])
(l,t,w,h)=rect
r=l+w#the coordinates are translated by 1 because of ihm zero pad
b=t+h
confidenceDict[rect].append(((ihm[b,r]+ihm[t,l])-(ihm[b,l]+ihm[t,r]))/(w*h))
res=np.array([tup[1]+(tup[0],) for tup in sorted([(max(confidenceDict[rec]),rec) for rec in confidenceDict.keys()],reverse=True)])
return res
def postProcessProp(propCSV):
lines=[tuple([float(c) for c in l.split(',')]) for l in propCSV.split('\n') if len(l)>0]
rectDict=defaultdict(list)
for l in lines:
rectDict[l[:4]].append(l[4:])
reslines=[]
for r in rectDict.keys():
reslines.append(r+max(rectDict[r]))
if len(reslines):
proposals=np.array(reslines)
proposals=proposals[np.argsort(-proposals[:,4]),:]
proposals=[[int(p[0]),int(p[1]),int(p[2]),int(p[3]),p[4],p[5],int(p[6]),int(p[7]),int(p[8]),int(p[9])] for p in proposals.tolist()]
return '\n'.join([','.join([str(c) for c in p]) for p in proposals])
else:
return ''
switches={'maxProposalsIoU':'200000',#IoU over this are not computed #20000
'gpu':None,
'img2propPath':'./tmp/TextProposalsInitialSuppression/img2hierarchy',
'weakClassifier':'./tmp/TextProposalsInitialSuppression/trained_boost_groups.xml',
'proto':'/tmp/dictnet_vgg_deploy.prototxt',
'pretrained':'/tmp/dictnet_vgg.caffemodel',
'vocabulary':'/tmp/dictnet_vgg_labels.txt',
'dictnetThr':'0.004',
'vocDir':'voc_strong',
'iouThr':'0.3',
'minibatchSz':'100',
'threads':'1',
'dontCareDictFile':'',
'IoUThresholds':'[.5]',
'extraPlotDirs':'{".":"Confidence"}',
'care':'True', #If true dont cares are supressed
'bayesianFname':'/tmp/bayesian.cPickle',
'plotter':'plt.semilogx',
'thr':'0.1',
'plotfname':'plots.pdf',
'fixedProps':'1000',#,
'propDir':'proposals'
}
if __name__=='__main__':
hlp="""
hm2conf blabla/*/heatma.../*.csv
hm2conf blabla/*/heatma.../*.png
img2prop blabla/*/input/*.jpg
"""
params=[(len(p)>0 and p[0]!='-',p) for p in sys.argv]
sys.argv=[p[1] for p in params if p[0]]
switches.update(dict([p[1][1:].split('=') for p in params if not p[0]]))
print 'Threads',int(switches['threads'])
if sys.argv[1]=='dictnet2final':
#dnThrCount=0
def worker(vggfname):
outDir='finalDN%dIou%dVoc%s_%s'%((int(100*eval(switches['dictnetThr']))),(int(100*eval(switches['iouThr']))),switches['vocDir'].split('_')[-1],vggfname.split('/')[-2])
ofname=vggfname.split('/')
ofname[-2]=outDir
ofname='/'.join(ofname)
LTWHConf,transcriptions=loadVggTranscription(vggfname)
#FIXING UNSORTED YOLO/TEXTBOXE BEGIN
LTWHTextnes=fname2Array(vggfname.replace('vggtr_',''))
LTWHTextnes[LTWHTextnes[:,2]<3,2]=3
LTWHTextnes[LTWHTextnes[:,3]<3,3]=3
if set([tuple(l) for l in LTWHTextnes[:,:4].astype('int32').tolist()])!=set([tuple(l) for l in LTWHConf[:,:4].astype('int32').tolist()]):
print "Count not fincd consistent conf "+vggfname
raise Exception()
sortdeterministic=lambda x: (x[:,0]+x[:,1]*(10^4)+x[:,0]*(10^8)+x[:,1]*(10^12)).argsort()
LTWHTextnes=LTWHTextnes[sortdeterministic(LTWHTextnes),:]#Making the textnes follow a dterministic order
idx=sortdeterministic(LTWHConf)#alligning vggtr with confidence rectangles
transcriptions=transcriptions[idx]#alligning vggtr with confidence rectangles
LTWHConf=LTWHConf[idx,:]#alligning vggtr with confidence rectangles
#now the 4 first columns of LTWHConf and LTWHTextnes should be the same rectangles
if LTWHTextnes[:,:4].astype('int32').tolist()!=LTWHConf[:,:4].astype('int32').tolist():
print "Failed to allign rectangles "+vggfname
raise Exception()
textnesSortedIdx=np.argsort(-LTWHTextnes[:,4])
LTWHConf=LTWHConf[textnesSortedIdx,:]
transcriptions=transcriptions[textnesSortedIdx]
#FIXING UNSORTED YOLO/TEXTBOXE END
print vggfname.split('/')[-2].split('_')[-1]+'/'+vggfname.split('/')[-1],' Initial :',LTWHConf.shape[0],
filterIdx=LTWHConf[:,4]>eval(switches['dictnetThr'])
print ' dictnet>'+str(float(int(eval(switches['dictnetThr'])*100))/100)+' kills:',(filterIdx==0).sum(),
LTWHConf=LTWHConf[filterIdx,:]
transcriptions=transcriptions[filterIdx]
#print LTWHConf[:,2:4].max()
#print '\n\n#2\n','\n'.join(transcriptions.tolist())
if switches['vocDir'] and LTWHConf.size>0:
vocFname=ofname.split('/')
vocFname[-2]=switches['vocDir']
vocFname='/'.join(vocFname)[:-3]+'txt'
voc=set([s.lower().strip() for s in myread(vocFname).split('\n')[:-1]])
filterIdx=np.ones(transcriptions.shape[0],dtype='bool')
for k in range(transcriptions.shape[0]):
filterIdx[k]=transcriptions[k].lower() in voc
print ' VOC kills:', (filterIdx==0).sum(),
LTWHConf=LTWHConf[filterIdx,:]
transcriptions=transcriptions[filterIdx]
res=[]
filterIdx=getNmsIdx(LTWHConf)
print 'NMS kills:', LTWHConf.shape[0]-(filterIdx).shape[0],' SURVIVED:',(filterIdx.shape[0])
LTWHConf=LTWHConf[filterIdx,:]
transcriptions=transcriptions[filterIdx]
for k in range(transcriptions.shape[0]):
resLine=','.join([str(int(c)) for c in LTWHConf[k,:4]])+','+transcriptions[k]
#print 'MAX RESLINE:',resLine
res.append(resLine)
open(ofname,'w').write('\n'.join(res))
outDirL= lambda x:'finalDN%dIou%dVoc%s_%s'%((int(100*eval(switches['dictnetThr']))),(int(100*eval(switches['iouThr']))),switches['vocDir'].split('_')[-1],x.split('/')[-2])
[go('mkdir -p '+d) for d in set('/'.join(f.split('/')[:-2]+[outDirL(f)]) for f in sys.argv[2:])]
if int(switches['threads'])<=1:
[worker(f) for f in sys.argv[2:]]
else:
pool=Pool(int(switches['threads']))
pool.map(worker,sys.argv[2:])
sys.exit(0)
if sys.argv[1]=='gt2SW':
widths=np.empty([1000000])
heights=np.empty([1000000])
objCount=0
for fname in sys.argv[2:]:
gtMat,transcriptions=loadTxtGtFile(fname)
widths[objCount:objCount+gtMat.shape[0]]=gtMat[:,2]
heights[objCount:objCount+gtMat.shape[0]]=gtMat[:,3]
objCount+=gtMat.shape[0]
widths=np.sort(widths[:objCount])[[objCount/6,2*objCount/6,3*objCount/6,4*objCount/6,5*objCount/6]]
heights=np.sort(heights[:objCount])[[objCount/6,2*objCount/6,3*objCount/6,4*objCount/6,5*objCount/6]]
#[go(cmd) for cmd in list(set([('mkdir -p '+('/'.join(f.split('/')[:-2]))+'/conf_rnd') for f in sys.argv[2]]))]
for fname in sys.argv[2:]:
ofname=fname.split('/')
ofname[-2]='conf_rnd'
cmd='mkdir -p '+'/'.join(ofname[:-1])
print cmd
go(cmd)
ofname='/'.join(ofname)[:-3]+'csv'
sz=(1280,720)
res=np.random.randint(0,100000000,[100000,5],dtype='int32')
res[:,2]=widths[res[:,2]%5]
res[:,3]=heights[res[:,3]%5]
res[:,0]=res[:,0]%(sz[0]-res[:,2])
res[:,1]=res[:,1]%(sz[1]-res[:,3])
print (sz[1]-res[:,1])
res=res.astype('float')
res[:,4]=np.random.rand(res.shape[0])
res=res[np.argsort(-res[:,4]),:]
open(ofname,'w').write('\n'.join([','.join([str(c) for c in l]) for l in res.tolist()]))
sys.exit(0)
if sys.argv[1]=='conf2IoU':
createRequiredDirs(sys.argv[2:],'+../iou_')
if switches['dontCareDictFile']!='':
dictionary=[l for l in open(switches['dontCareDictFile']).read().split('\n') if len(l)]
else:
dictionary=[]
def worker(confFname):
t=time.time()
confMat=fname2Array(confFname)
#print confMat.shape
if confMat.shape[0] == 5:
confMat = np.array([confMat])
#print confMat.shape
idx=np.argsort(-confMat[:,4]) #For others (Suppression, FCN)
#idx=np.argsort(confMat[:,4]) #For TP
confMat=confMat[idx,:]
#confMat=confMat[reversed(idx),:]#UGLY!!!!!!!!!!!!!!!!!!!!!!!!!!
try:
gtMat,transcriptions=loadTxtGtFile(getGtFromConf(confFname))
except:
open("/tmp/blacklist.txt","a").write(confFname.split("/")[-1] + '\n')
lines=[[' '.join(c.split()) for c in l.split(",") ] for l in open(getGtFromConf(confFname)).read().split() if len(l.split(","))==5]
cleanLines=[]
cleanTranscriptions=[]
for line in lines:
try:
cleanLines.append([float(c) for c in line[:4]] )
cleanTranscriptions.append(line[-1])
except:
pass
gtMat=np.array(cleanLines)
transcriptions=np.empty(gtMat.shape[0],dtype=object)
transcriptions[:]=cleanTranscriptions
#except:
# print "BADGT: ",confFname
dontCare=getDontCare(transcriptions,dictionary)
(IoU,I,U,minsurf)=get2PointIU(gtMat,confMat)
augmentedIoU=np.empty([IoU.shape[0]+1,IoU.shape[1]])
augmentedIoU[:-1,:]=IoU
augmentedIoU[-1,:]=dontCare
#print 'Transcriptions: ',transcriptions
#print 'DONT CARE: ',dontCare
#array2csvFname(IoU,getIouFromConf(confFname))
array2pngFname(augmentedIoU,getIouFromConf(confFname))
#open(getIouFromConf(confFname),'w').write(arrayToCsvStr(IoU))
print confFname, ' to ', getIouFromConf(confFname), ' ', int((time.time()-t)*1000)/1000.0,' sec.'
return None
pool=Pool(int(switches['threads']))
if eval(switches['threads'])==1:
[worker(f) for f in sys.argv[2:]]
else:
pool=Pool(int(switches['threads']))
pool.map(worker,sys.argv[2:])
sys.exit(0)
if sys.argv[1]=='prop2conf':
def worker(propFname):
proposals=fname2Array(propFname)
propRects=proposals[:,:5]#.astype('int32')
confidenceDict=defaultdict(list)
for rectId in range(proposals.shape[0]):
rect=tuple(propRects[rectId,:])
confidenceDict[rect].append(proposals[rectId,4])
newProds=reversed(sorted([(max(confidenceDict[rec]),rec) for rec in confidenceDict.keys()]))
arr=np.array([list(s[1])+list([s[0]]) for s in newProds],dtype='float')
array2csvFname(arr,getConfidenseFromProposal(propFname))
createRequiredDirs(sys.argv[2:],'+../conf_')
pool=Pool(int(switches['threads']))
if eval(switches['threads'])==1:
[worker(f) for f in sys.argv[2:]]
else:
pool=Pool(int(switches['threads']))
pool.map(worker,sys.argv[2:])
sys.exit(0)
if sys.argv[1]=='getCumRecall':
def getConfFromConf(fname,confName):
if confName=='.':
return fname
else:
return '/'.join(fname.split('/')[:-2]+[confName]+fname.split('/')[-1:])
iouThr=eval(switches['IoUThresholds'])
maxProposals=eval(switches['maxProposalsIoU'])
#resGtObjDetected=np.zeros([maxProposals,len(iouThr)],dtype='int64')
resGtObjCount=0
confDict=eval(switches['extraPlotDirs'])
resGtObjDetected=dict([(k,np.zeros([maxProposals,len(iouThr)],dtype='int64')) for k in confDict.keys()])
for confFname in sys.argv[2:]:
for confStr in confDict.keys():
#print 'Conf ',getConfFromConf(confFname,confStr)
#print 'IoU ',getIouFromConf(getConfFromConf(confFname,confStr))
try:
augmentedIoU=fname2Array(getIouFromConf(getConfFromConf(confFname,confStr)))
except:
print 'Could not load ',getIouFromConf(getConfFromConf(confFname,confStr))
raise Exception()
if augmentedIoU.shape[0]>1:
care=augmentedIoU[-1,:]
keepProposals=min(maxProposals,augmentedIoU.shape[0]-1)
if eval(switches['care']):
IoU=augmentedIoU[:keepProposals,care.astype('bool')]
else:
IoU=augmentedIoU[:keepProposals,:]
#print care
#print IoU.shape
resGtObjCount+=IoU.shape[1]
#dontCare=augmentedIoU[:-1,:]
#IoU[:,np.argmax(IoU,axis=1)]*=2#removin non maximal matches (a proposal identifies at most an object)
#IoU-=augmentedIoU[:keepProposals,:]
for tNum in range(len(iouThr)):
thr=iouThr[tNum]
found=((IoU>=thr).cumsum(axis=0)>0)
resGtObjDetected[confStr][:keepProposals,tNum]+=found.sum(axis=1)
resGtObjDetected[confStr][keepProposals:,tNum]+=found.sum(axis=1).max()
#print IoU[:3,:]
sortedKeysByLegends=[e[1][0] for e in sorted([(confDict[k],(k,confDict[k])) for k in confDict.keys()])];colors="rgbcmyk"*10
count=0
for confStr in sortedKeysByLegends:
#for confStr in confDict.keys():
plt.plot(resGtObjDetected[confStr].astype('float')/(resGtObjCount/len(confDict.keys())), colors[count] ,label=confDict[confStr])
count+=1
#plt.legend(confDict[confStr])
plt.ylim(0,1)
plt.xscale('log')
plt.xlabel('# of proposals')
plt.ylabel('Detection Rate')
plt.legend(loc='upper left')
#plt.title('COCO-Text')
#plt.title('ICDAR')
plt.grid()
plt.savefig(switches['plotfname'])
sys.exit(0)