东方耀AI技术分享
标题:
总结:SSD预测输出后的解码过程
[打印本页]
作者:
东方耀
时间:
2019-12-2 18:05
标题:
总结:SSD预测输出后的解码过程
总结:SSD预测输出后的解码过程:
confidence_thresh=0.6,
iou_threshold=0.45,
top_k=100,
1、拿到模型预测后输出的矩阵
2、一系列变换:边界框boxes坐标由相对值变为绝对值,坐标形式centroids2corners,反归一化
3、一张图片一张图片的处理 一种类别一种类别的处理
4、同一个类别中,预测概率值经过给定置信度阈值的过滤后取概率最大的box
5、用剩下的box分别与概率最大的box计算iou值
6、小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)
7、循环往复,直到剩下的box个数为0,最终NMS我们需要的结果是:每一次的选出的概率最大的box
8、在0维上加上对应的类别id,即可得到所有检测到的物体的类别id和定位坐标box
9、如果数量比较多,则需要经过top-k过滤
原始预测值 y_pred: (6, 17538, 18)
维度截断前10,预测boxes的相对值变绝对值,坐标(centroids)中心点变成角形式(xmin,ymin,xmax,ymax),反归一化变成实际像素值 shape= (6, 17538, 10)
正在处理的图片编号:0
在17538个box中(先都看同一个类别)类别为:1的概率值大于给定阈值的box个数:(11, 5)
threshold_met: [[ 0.96580559 33.61340046 150.98859072 136.08298302 210.48692465]
[ 0.97073776 35.04751325 153.48874927 141.03665829 205.3978622 ]
[ 0.95259297 39.73779559 149.55267012 138.34213257 206.87343478]
[ 0.89510643 39.29663658 151.50745511 132.8565073 211.78327203]
[ 0.94443005 37.43476868 151.41758323 134.86936569 205.93265891]
[ 0.79319638 40.46634436 147.42987156 128.49414825 202.48596668]
[ 0.96809524 27.92355537 155.43519258 136.28113747 214.9014473 ]
[ 0.98907334 24.69376087 150.93723536 132.63394833 204.56310511]
[ 0.95623392 22.58502245 147.32278883 131.53581619 208.34618211]
[ 0.9804498 22.16494203 152.472049 132.7502346 209.70026851]
[ 0.94763917 29.06545758 149.00491834 134.20828342 210.49881577]]
先得到同一个类别中,概率值最大的box: [ 0.98907334 24.69376087 150.93723536 132.63394833 204.56310511]
用10个剩下的box分别与概率最大的box计算iou值:similarities.shape=(10,)
iou的具体值(mode=element-wise): [0.8062058 0.79038136 0.77129147 0.7647146 0.83756321 0.74423871
0.72578245 0.85420101 0.86579716 0.82893815]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (0, 5)
maxima(经过NMS后的结果): (1, 5) [[ 0.98907334 24.69376087 150.93723536 132.63394833 204.56310511]]
maxima_output(在0维上加上对应的类别id): (1, 6) [[ 1. 0.98907334 24.69376087 150.93723536 132.63394833
204.56310511]]
在17538个box中(先都看同一个类别)类别为:2的概率值大于给定阈值的box个数:(1, 5)
threshold_met: [[ 0.61496943 215.41189671 111.31170094 378.15725327 217.35234261]]
先得到同一个类别中,概率值最大的box: [ 0.61496943 215.41189671 111.31170094 378.15725327 217.35234261]
maxima(经过NMS后的结果): (1, 5) [[ 0.61496943 215.41189671 111.31170094 378.15725327 217.35234261]]
maxima_output(在0维上加上对应的类别id): (1, 6) [[ 2. 0.61496943 215.41189671 111.31170094 378.15725327
217.35234261]]
在17538个box中(先都看同一个类别)类别为:3的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:4的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:5的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
所有正样本类别的就处理完了pred= (2, 6) [[ 1. 0.98907334 24.69376087 150.93723536 132.63394833
204.56310511]
[ 2. 0.61496943 215.41189671 111.31170094 378.15725327
217.35234261]]
正在处理的图片编号:1
在17538个box中(先都看同一个类别)类别为:1的概率值大于给定阈值的box个数:(77, 5)
threshold_met: [[ 0.72249418 219.75242615 153.1521678 241.30391121 172.38067389]
[ 0.63175714 224.0515995 153.88339162 247.74650574 172.52631783]
[ 0.78431815 49.97917414 156.60817623 95.96534014 191.85233116]
[ 0.97783732 52.20789671 158.34558606 100.47534227 190.98113179]
[ 0.66519636 60.76698303 157.64154196 105.52818775 185.77827215]
[ 0.83306515 75.56817055 156.78489804 113.13670635 181.3839376 ]
[ 0.76062477 82.64760017 158.90575647 114.19803143 183.16969872]
[ 0.9681372 81.52966976 158.95491242 118.49089622 181.38511777]
[ 0.72419327 75.96833467 157.8263998 119.00913477 182.48233795]
[ 0.88565707 87.54989862 158.64579678 124.78005409 181.21948242]
[ 0.76149821 85.91345787 157.34673142 129.07217503 181.88368678]
[ 0.91055995 105.39467812 159.93198752 130.98977566 179.53603864]
[ 0.93923956 101.22835636 158.29587579 136.55478001 178.74550223]
[ 0.77576023 97.51484871 157.45096207 137.86335468 177.55844593]
[ 0.9875946 113.33487511 158.25378299 137.9722023 176.57663226]
[ 0.99766099 112.83602715 157.94686675 139.12671089 176.06701255]
[ 0.98388278 108.79235744 156.75501823 139.50529575 176.702106 ]
[ 0.61978006 124.07536983 158.51563811 138.35942745 175.7309854 ]
[ 0.99592662 121.07721806 158.33714604 142.34191418 174.59809184]
[ 0.99863428 119.32604313 158.375144 143.32727909 174.77560043]
[ 0.98871082 118.16326618 156.49112463 145.52780628 174.63476658]
[ 0.96136111 130.29853821 158.1648767 150.66467285 172.93472886]
[ 0.9956851 129.40013409 157.70614743 152.63318539 173.20198417]
[ 0.97767442 127.52254486 156.90146685 153.92612457 172.85417318]
[ 0.84387225 136.15815639 158.35917592 153.13718319 171.19165063]
[ 0.89749861 133.18143368 158.85887146 153.54702473 171.89326286]
[ 0.70893997 133.25424671 158.37911367 154.75643635 171.28146887]
[ 0.61332923 218.48110199 154.73957062 240.66309929 179.40852642]
[ 0.9830699 221.70513153 156.12831116 242.03518867 175.49525499]
[ 0.90813005 222.41485119 155.08598685 244.20275688 174.52545762]
[ 0.62235928 226.32760048 156.82033896 245.16139984 179.26374078]
[ 0.98147362 224.30598736 157.73985386 244.48144913 176.41382217]
[ 0.74661607 224.83150005 155.18934131 250.34960747 178.35654616]
[ 0.9671545 222.86640644 155.90136051 245.54406166 177.41868496]
[ 0.61027837 234.94965076 156.99382424 256.05279922 185.51986814]
[ 0.81136084 232.95317173 157.17038512 260.53679466 184.45406556]
[ 0.8977809 237.0672369 156.65000081 269.07259941 187.75847554]
[ 0.6113261 52.17610359 157.48654604 88.91575813 194.47270632]
[ 0.69243836 53.53913784 160.01324058 94.62131023 190.34877419]
[ 0.99926549 50.32630205 160.6002152 99.16625261 190.06789327]
[ 0.63045138 57.20763803 161.08161807 93.30059052 188.86508346]
[ 0.94717294 46.96068048 159.42901969 104.41734552 191.21856093]
[ 0.99991632 48.14987183 162.33640909 99.23004627 189.70298767]
[ 0.9107694 48.7369895 160.88706851 100.56310415 190.16273618]
[ 0.66317242 47.47363329 156.22842908 105.59125185 189.37366605]
[ 0.98607868 50.74757695 159.10068154 98.67894173 185.75114608]
[ 0.82275349 66.52825356 157.45414495 107.58130074 182.43995905]
[ 0.83345973 79.20886517 157.92124271 117.09488869 180.25091887]
[ 0.65916258 85.87562799 158.82571936 123.02241325 181.4479351 ]
[ 0.68752068 233.41815948 155.51054478 261.57938004 188.57127428]
[ 0.88525099 235.58012009 156.85658455 264.50099945 190.96859694]
[ 0.78732401 244.17614937 157.38059878 275.00201225 193.05340648]
[ 0.98235148 50.27100563 160.94330549 95.57601929 192.66947508]
[ 0.99753368 49.24220324 159.10289884 93.05137396 187.49179244]
[ 0.66598719 43.60121369 159.54807401 96.89675331 191.14750028]
[ 0.99362344 53.05385113 162.62837648 100.8002615 189.69715834]
[ 0.99742162 53.54989529 165.17379284 103.67035389 189.95858431]
[ 0.98410797 53.69304657 163.722682 101.04702473 190.5049324 ]
[ 0.85896868 58.69515181 159.42880511 101.96716547 185.66706777]
[ 0.99778509 49.81727958 162.8572762 97.03897476 186.05598807]
[ 0.99320203 46.62781477 162.64715195 91.62993193 188.89600039]
[ 0.91097564 242.31081963 153.76625061 287.04068184 194.19704676]
[ 0.67423779 242.34140396 153.26884389 287.0425415 195.27474046]
[ 0.79372537 252.5521946 150.29686689 302.56052971 202.82721519]
[ 0.67557091 251.07719421 152.27222443 301.91019058 198.49802256]
[ 0.60347539 262.34584808 140.18645883 330.25497437 212.12155223]
[ 0.93310475 278.59125137 133.46484303 406.14732742 240.08486867]
[ 0.88905698 285.05822182 125.94563663 414.32135582 232.7352047 ]
[ 0.94346797 273.73829842 140.49496651 405.59214592 228.25380564]
[ 0.70535594 282.78608322 144.89312768 411.64209366 227.63848901]
[ 0.65507102 287.64707565 141.68227315 432.68651962 241.3729012 ]
[ 0.88083708 285.42766571 138.64716589 441.9786644 246.06478214]
[ 0.85064954 283.10666084 145.36702037 434.47663307 231.66444898]
[ 0.73932117 291.0688591 145.39954662 454.62867737 230.73506355]
[ 0.69014585 296.09524727 130.34936786 417.02490807 237.44975924]
[ 0.86166078 302.17955589 138.09015155 450.4754734 251.24793649]
[ 0.71307981 311.71751976 139.75530267 475.39721489 269.67009902]]
先得到同一个类别中,概率值最大的box: [ 0.99991632 48.14987183 162.33640909 99.23004627 189.70298767]
用76个剩下的box分别与概率最大的box计算iou值:similarities.shape=(76,)
iou的具体值(mode=element-wise): [0. 0. 0.71498715 0.76311688 0.51356002 0.24084369
0.19002992 0.17841835 0.235436 0.10930819 0.11850432 0.
0. 0.01195957 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0.5741067 0.73972468 0.8907537 0.66346473 0.76532905
0.90120122 0.71073877 0.72267417 0.37222167 0.19025522 0.12870508
0. 0. 0. 0.77713582 0.71581866 0.7631944
0.86777219 0.73735489 0.79771335 0.59569531 0.78366882 0.79376011
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (56, 5)
先得到同一个类别中,概率值最大的box: [ 0.99863428 119.32604313 158.375144 143.32727909 174.77560043]
用55个剩下的box分别与概率最大的box计算iou值:similarities.shape=(55,)
iou的具体值(mode=element-wise): [0. 0. 0. 0. 0. 0.
0.07675909 0.12364191 0.23970631 0.33900959 0.33745629 0.56709233
0.59554475 0.49006536 0.57026954 0.87459872 0.78070099 0.3758211
0.37740123 0.39050547 0.17682375 0.25101563 0.24019076 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.05017515
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (49, 5)
先得到同一个类别中,概率值最大的box: [ 0.9956851 129.40013409 157.70614743 152.63318539 173.20198417]
用48个剩下的box分别与概率最大的box计算iou值:similarities.shape=(48,)
iou的具体值(mode=element-wise): [0. 0. 0. 0. 0. 0.
0. 0. 0.02509198 0.10929508 0.1260791 0.83553275
0.8198082 0.57687623 0.68169777 0.64539796 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (43, 5)
先得到同一个类别中,概率值最大的box: [ 0.9830699 221.70513153 156.12831116 242.03518867 175.49525499]
用42个剩下的box分别与概率最大的box计算iou值:similarities.shape=(42,)
iou的具体值(mode=element-wise): [0.65057521 0.54551752 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.63989328
0.79102305 0.56078146 0.69078454 0.51122726 0.72727727 0.15161736
0.16984282 0.07225356 0. 0. 0. 0.14413073
0.09548975 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (34, 5)
先得到同一个类别中,概率值最大的box: [ 0.9681372 81.52966976 158.95491242 118.49089622 181.38511777]
用33个剩下的box分别与概率最大的box计算iou值:similarities.shape=(33,)
iou的具体值(mode=element-wise): [0.6788535 0.79792557 0.78122792 0.70252993 0.63139252 0.23903224
0.2823847 0.31215136 0. 0. 0. 0.45995059
0.82538095 0.78006928 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (25, 5)
先得到同一个类别中,概率值最大的box: [ 0.94346797 273.73829842 140.49496651 405.59214592 228.25380564]
用24个剩下的box分别与概率最大的box计算iou值:similarities.shape=(24,)
iou的具体值(mode=element-wise): [0. 0. 0. 0. 0. 0.
0. 0. 0.00357045 0.0418803 0.04335517 0.11936389
0.10320008 0.3262412 0.79463235 0.7148508 0.84175594 0.64544
0.59104297 0.70104972 0.59156524 0.64433127 0.47079207 0.33492643]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (15, 5)
先得到同一个类别中,概率值最大的box: [ 0.93923956 101.22835636 158.29587579 136.55478001 178.74550223]
用14个剩下的box分别与概率最大的box计算iou值:similarities.shape=(14,)
iou的具体值(mode=element-wise): [0.64840278 0.79752075 0. 0. 0. 0.
0. 0. 0. 0. 0. 0.
0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (12, 5)
先得到同一个类别中,概率值最大的box: [ 0.91097564 242.31081963 153.76625061 287.04068184 194.19704676]
用11个剩下的box分别与概率最大的box计算iou值:similarities.shape=(11,)
iou的具体值(mode=element-wise): [0.19421066 0.24095247 0.42225857 0.3029937 0.37140765 0.60805474
0.96182943 0.45852904 0.53768868 0.17531475 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (7, 5)
先得到同一个类别中,概率值最大的box: [ 0.8977809 237.0672369 156.65000081 269.07259941 187.75847554]
用6个剩下的box分别与概率最大的box计算iou值:similarities.shape=(6,)
iou的具体值(mode=element-wise): [0.51284113 0.57798057 0.65502422 0.74729463 0.03689705 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (2, 5)
先得到同一个类别中,概率值最大的box: [ 0.71307981 311.71751976 139.75530267 475.39721489 269.67009902]
用1个剩下的box分别与概率最大的box计算iou值:similarities.shape=(1,)
iou的具体值(mode=element-wise): [0.0537353]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (1, 5)
先得到同一个类别中,概率值最大的box: [ 0.60347539 262.34584808 140.18645883 330.25497437 212.12155223]
maxima(经过NMS后的结果): (11, 5) [[ 0.99991632 48.14987183 162.33640909 99.23004627 189.70298767]
[ 0.99863428 119.32604313 158.375144 143.32727909 174.77560043]
[ 0.9956851 129.40013409 157.70614743 152.63318539 173.20198417]
[ 0.9830699 221.70513153 156.12831116 242.03518867 175.49525499]
[ 0.9681372 81.52966976 158.95491242 118.49089622 181.38511777]
[ 0.94346797 273.73829842 140.49496651 405.59214592 228.25380564]
[ 0.93923956 101.22835636 158.29587579 136.55478001 178.74550223]
[ 0.91097564 242.31081963 153.76625061 287.04068184 194.19704676]
[ 0.8977809 237.0672369 156.65000081 269.07259941 187.75847554]
[ 0.71307981 311.71751976 139.75530267 475.39721489 269.67009902]
[ 0.60347539 262.34584808 140.18645883 330.25497437 212.12155223]]
maxima_output(在0维上加上对应的类别id): (11, 6) [[ 1. 0.99991632 48.14987183 162.33640909 99.23004627
189.70298767]
[ 1. 0.99863428 119.32604313 158.375144 143.32727909
174.77560043]
[ 1. 0.9956851 129.40013409 157.70614743 152.63318539
173.20198417]
[ 1. 0.9830699 221.70513153 156.12831116 242.03518867
175.49525499]
[ 1. 0.9681372 81.52966976 158.95491242 118.49089622
181.38511777]
[ 1. 0.94346797 273.73829842 140.49496651 405.59214592
228.25380564]
[ 1. 0.93923956 101.22835636 158.29587579 136.55478001
178.74550223]
[ 1. 0.91097564 242.31081963 153.76625061 287.04068184
194.19704676]
[ 1. 0.8977809 237.0672369 156.65000081 269.07259941
187.75847554]
[ 1. 0.71307981 311.71751976 139.75530267 475.39721489
269.67009902]
[ 1. 0.60347539 262.34584808 140.18645883 330.25497437
212.12155223]]
在17538个box中(先都看同一个类别)类别为:2的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:3的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:4的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:5的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
所有正样本类别的就处理完了pred= (11, 6) [[ 1. 0.99991632 48.14987183 162.33640909 99.23004627
189.70298767]
[ 1. 0.99863428 119.32604313 158.375144 143.32727909
174.77560043]
[ 1. 0.9956851 129.40013409 157.70614743 152.63318539
173.20198417]
[ 1. 0.9830699 221.70513153 156.12831116 242.03518867
175.49525499]
[ 1. 0.9681372 81.52966976 158.95491242 118.49089622
181.38511777]
[ 1. 0.94346797 273.73829842 140.49496651 405.59214592
228.25380564]
[ 1. 0.93923956 101.22835636 158.29587579 136.55478001
178.74550223]
[ 1. 0.91097564 242.31081963 153.76625061 287.04068184
194.19704676]
[ 1. 0.8977809 237.0672369 156.65000081 269.07259941
187.75847554]
[ 1. 0.71307981 311.71751976 139.75530267 475.39721489
269.67009902]
[ 1. 0.60347539 262.34584808 140.18645883 330.25497437
212.12155223]]
正在处理的图片编号:2
在17538个box中(先都看同一个类别)类别为:1的概率值大于给定阈值的box个数:(9, 5)
threshold_met: [[ 0.75215954 126.81917667 150.52061677 166.79785252 174.54941869]
[ 0.72137636 118.82426262 156.49695396 148.21206093 173.96403551]
[ 0.62573338 127.13476181 155.17378449 155.91001511 172.95268178]
[ 0.8579843 122.28881836 155.35283089 159.26702499 174.06102419]
[ 0.77935374 117.39601851 153.4496069 161.96522713 173.70997667]
[ 0.69708437 132.20377922 154.79691625 160.48350334 173.55682254]
[ 0.76109886 125.37506104 152.20177174 166.21584892 176.1164546 ]
[ 0.84214813 128.04242134 155.31349182 162.44550705 174.37398434]
[ 0.71698338 120.72495461 152.64616013 164.87989426 174.33693409]]
先得到同一个类别中,概率值最大的box: [ 0.8579843 122.28881836 155.35283089 159.26702499 174.06102419]
用8个剩下的box分别与概率最大的box计算iou值:similarities.shape=(8,)
iou的具体值(mode=element-wise): [0.5806809 0.60188316 0.72665464 0.74108815 0.67518874 0.61295194
0.76522371 0.72230954]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (0, 5)
maxima(经过NMS后的结果): (1, 5) [[ 0.8579843 122.28881836 155.35283089 159.26702499 174.06102419]]
maxima_output(在0维上加上对应的类别id): (1, 6) [[ 1. 0.8579843 122.28881836 155.35283089 159.26702499
174.06102419]]
在17538个box中(先都看同一个类别)类别为:2的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:3的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:4的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:5的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
所有正样本类别的就处理完了pred= (1, 6) [[ 1. 0.8579843 122.28881836 155.35283089 159.26702499
174.06102419]]
正在处理的图片编号:3
在17538个box中(先都看同一个类别)类别为:1的概率值大于给定阈值的box个数:(16, 5)
threshold_met: [[ 0.83615065 31.17763281 131.31062686 181.47391319 253.48488092]
[ 0.98970848 37.51424789 125.80279112 209.18792725 267.2634244 ]
[ 0.93533438 46.49530649 115.87157249 225.3269577 268.55342388]
[ 0.8587783 27.37287283 136.90109253 173.07815552 266.76249504]
[ 0.99867141 34.14845467 136.68976128 206.80286407 271.14822865]
[ 0.99361199 44.92258072 126.22027695 224.30296898 275.00020266]
[ 0.65652001 31.03308678 154.25248146 203.0613327 273.74925613]
[ 0.83881658 46.58774614 151.75019503 206.27194405 295.52804232]
[ 0.99856502 38.07931423 133.35420191 218.76410007 267.36681461]
[ 0.95636439 35.37297964 135.80897748 214.43787575 274.3303299 ]
[ 0.99495357 28.24527025 141.43128991 220.69158554 269.15100217]
[ 0.88248748 18.37405443 126.62755251 231.09898567 268.17290783]
[ 0.83220071 39.43939447 118.30216348 220.38265228 269.59469318]
[ 0.98419178 25.70425987 124.55210388 225.48322678 264.49656487]
[ 0.93223453 -6.6250205 126.96025372 240.31868935 273.29034805]
[ 0.6840108 26.47614956 143.79932284 220.59578419 280.51208854]]
先得到同一个类别中,概率值最大的box: [ 0.99867141 34.14845467 136.68976128 206.80286407 271.14822865]
用15个剩下的box分别与概率最大的box计算iou值:similarities.shape=(15,)
iou的具体值(mode=element-wise): [0.70606035 0.87042296 0.71948317 0.74877797 0.77358186 0.82181013
0.70333616 0.86870696 0.92315419 0.85660799 0.74128782 0.78468985
0.75812911 0.64244242 0.79187667]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (0, 5)
maxima(经过NMS后的结果): (1, 5) [[ 0.99867141 34.14845467 136.68976128 206.80286407 271.14822865]]
maxima_output(在0维上加上对应的类别id): (1, 6) [[ 1. 0.99867141 34.14845467 136.68976128 206.80286407
271.14822865]]
在17538个box中(先都看同一个类别)类别为:2的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:3的概率值大于给定阈值的box个数:(2, 5)
threshold_met: [[ 0.62705225 276.77527428 123.70916605 323.44805717 211.40727997]
[ 0.77580345 273.19664955 136.21562719 321.30354881 213.64070177]]
先得到同一个类别中,概率值最大的box: [ 0.77580345 273.19664955 136.21562719 321.30354881 213.64070177]
用1个剩下的box分别与概率最大的box计算iou值:similarities.shape=(1,)
iou的具体值(mode=element-wise): [0.74908807]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (0, 5)
maxima(经过NMS后的结果): (1, 5) [[ 0.77580345 273.19664955 136.21562719 321.30354881 213.64070177]]
maxima_output(在0维上加上对应的类别id): (1, 6) [[ 3. 0.77580345 273.19664955 136.21562719 321.30354881
213.64070177]]
在17538个box中(先都看同一个类别)类别为:4的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:5的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
所有正样本类别的就处理完了pred= (2, 6) [[ 1. 0.99867141 34.14845467 136.68976128 206.80286407
271.14822865]
[ 3. 0.77580345 273.19664955 136.21562719 321.30354881
213.64070177]]
正在处理的图片编号:4
在17538个box中(先都看同一个类别)类别为:1的概率值大于给定阈值的box个数:(9, 5)
threshold_met: [[ 0.89631104 159.06984329 141.68805778 187.76189804 165.12234807]
[ 0.90466708 155.8009243 141.27936065 188.0476284 165.78345895]
[ 0.81128764 158.84270668 141.43401682 188.05990219 164.74363804]
[ 0.95014256 159.748106 146.04595006 186.57394409 166.58906937]
[ 0.78325415 156.25057697 143.61928403 190.64013004 167.08084345]
[ 0.95104986 156.92095757 146.14132941 187.00819016 166.46257639]
[ 0.76541919 153.07462692 146.18838429 187.50088692 166.10714793]
[ 0.69924593 161.54718876 144.94376779 186.80349827 166.49824977]
[ 0.75742286 159.24526691 146.20946646 184.93296146 166.13563299]]
先得到同一个类别中,概率值最大的box: [ 0.95104986 156.92095757 146.14132941 187.00819016 166.46257639]
用8个剩下的box分别与概率最大的box计算iou值:similarities.shape=(8,)
iou的具体值(mode=element-wise): [0.70378965 0.72905334 0.68176577 0.88300622 0.75779056 0.85879581
0.79874574 0.83717507]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (0, 5)
maxima(经过NMS后的结果): (1, 5) [[ 0.95104986 156.92095757 146.14132941 187.00819016 166.46257639]]
maxima_output(在0维上加上对应的类别id): (1, 6) [[ 1. 0.95104986 156.92095757 146.14132941 187.00819016
166.46257639]]
在17538个box中(先都看同一个类别)类别为:2的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:3的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:4的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:5的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
所有正样本类别的就处理完了pred= (1, 6) [[ 1. 0.95104986 156.92095757 146.14132941 187.00819016
166.46257639]]
正在处理的图片编号:5
在17538个box中(先都看同一个类别)类别为:1的概率值大于给定阈值的box个数:(33, 5)
threshold_met: [[ 0.79548252 141.32047176 150.02050996 162.78303623 167.98333526]
[ 0.85634446 141.29078865 150.84210634 164.34156418 167.43335724]
[ 0.86653131 141.7171669 150.04974604 165.22047043 167.55713224]
[ 0.87877679 140.23023605 150.45549273 165.6980896 167.42046475]
[ 0.6583057 140.80884933 150.55536032 167.6718235 166.83247089]
[ 0.78633976 177.22223282 149.77755547 204.15175438 172.92995453]
[ 0.938685 178.00370693 151.63866878 209.14591312 176.80408359]
[ 0.64307755 175.72311401 149.79396164 205.59270859 171.76904082]
[ 0.95129114 179.99699593 150.00847578 211.26763344 175.18122196]
[ 0.99564457 179.13518429 151.96942091 212.69395351 176.72950029]
[ 0.90695029 179.63258743 150.6311059 210.85103989 174.56452847]
[ 0.88417858 269.14887428 150.45765638 306.57797813 176.9205451 ]
[ 0.78744322 177.78748512 152.63578892 202.94514656 178.33904028]
[ 0.99593568 177.95637131 152.51301527 207.7253437 176.57525539]
[ 0.99452776 175.70822239 152.81785727 207.99086094 179.46288586]
[ 0.99065512 178.68970871 152.64905691 207.49136925 176.89304352]
[ 0.83494103 176.19021893 151.80852413 208.4480238 177.04886198]
[ 0.93428355 179.28159714 150.00302196 208.25065613 178.76277566]
[ 0.99940383 177.28039742 151.95738673 210.60796738 176.8543303 ]
[ 0.99863523 178.41635227 151.37447119 211.8632555 177.7857542 ]
[ 0.99869847 176.68629169 151.8278718 210.20323277 177.48749256]
[ 0.94521946 171.44990444 149.40937757 213.64281178 177.55147219]
[ 0.71909463 263.66294861 152.94653177 301.93222046 180.97071648]
[ 0.96401632 269.98397827 153.22329998 300.86242676 177.0947814 ]
[ 0.9752388 266.84000015 152.23551393 305.36424637 179.28866744]
[ 0.94146687 267.47202873 153.76878977 301.30379677 177.80191898]
[ 0.6493988 259.9168396 153.13982964 304.83364105 178.02529335]
[ 0.64769369 271.7250824 151.85735822 304.10453796 176.82393193]
[ 0.87841833 269.64998245 149.27762747 307.32902527 177.65954733]
[ 0.86343527 176.38857365 151.91116333 210.36762714 179.45466042]
[ 0.76632452 265.59811592 154.0679276 302.19964027 179.14815545]
[ 0.80384678 347.57612228 137.46753931 464.33950424 223.73067141]
[ 0.71599853 341.07370377 138.16979527 463.92436981 217.09472537]]
先得到同一个类别中,概率值最大的box: [ 0.99940383 177.28039742 151.95738673 210.60796738 176.8543303 ]
用32个剩下的box分别与概率最大的box计算iou值:similarities.shape=(32,)
iou的具体值(mode=element-wise): [0. 0. 0. 0. 0. 0.63344539
0.92151863 0.60624402 0.78464658 0.88412212 0.79879247 0.
0.70266097 0.86327709 0.77558936 0.83906312 0.8940115 0.76592786
0.87913815 0.94204094 0.69880148 0. 0. 0.
0. 0. 0. 0. 0.87459752 0.
0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (16, 5)
先得到同一个类别中,概率值最大的box: [ 0.9752388 266.84000015 152.23551393 305.36424637 179.28866744]
用15个剩下的box分别与概率最大的box计算iou值:similarities.shape=(15,)
iou的具体值(mode=element-wise): [0. 0. 0. 0. 0. 0.78508134
0.77663816 0.70726602 0.78015872 0.77850835 0.75504954 0.75440068
0.82622498 0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (7, 5)
先得到同一个类别中,概率值最大的box: [ 0.87877679 140.23023605 150.45549273 165.6980896 167.42046475]
用6个剩下的box分别与概率最大的box计算iou值:similarities.shape=(6,)
iou的具体值(mode=element-wise): [0.80293173 0.88385894 0.8964118 0.87276022 0. 0. ]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (2, 5)
先得到同一个类别中,概率值最大的box: [ 0.80384678 347.57612228 137.46753931 464.33950424 223.73067141]
用1个剩下的box分别与概率最大的box计算iou值:similarities.shape=(1,)
iou的具体值(mode=element-wise): [0.86748001]
小于iou阈值的则保留(大于的表示与概率最大的box重合度非常高 需要丢弃)boxes: (0, 5)
maxima(经过NMS后的结果): (4, 5) [[ 0.99940383 177.28039742 151.95738673 210.60796738 176.8543303 ]
[ 0.9752388 266.84000015 152.23551393 305.36424637 179.28866744]
[ 0.87877679 140.23023605 150.45549273 165.6980896 167.42046475]
[ 0.80384678 347.57612228 137.46753931 464.33950424 223.73067141]]
maxima_output(在0维上加上对应的类别id): (4, 6) [[ 1. 0.99940383 177.28039742 151.95738673 210.60796738
176.8543303 ]
[ 1. 0.9752388 266.84000015 152.23551393 305.36424637
179.28866744]
[ 1. 0.87877679 140.23023605 150.45549273 165.6980896
167.42046475]
[ 1. 0.80384678 347.57612228 137.46753931 464.33950424
223.73067141]]
在17538个box中(先都看同一个类别)类别为:2的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:3的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:4的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
在17538个box中(先都看同一个类别)类别为:5的概率值大于给定阈值的box个数:(0, 5)
threshold_met: []
所有正样本类别的就处理完了pred= (4, 6) [[ 1. 0.99940383 177.28039742 151.95738673 210.60796738
176.8543303 ]
[ 1. 0.9752388 266.84000015 152.23551393 305.36424637
179.28866744]
[ 1. 0.87877679 140.23023605 150.45549273 165.6980896
167.42046475]
[ 1. 0.80384678 347.57612228 137.46753931 464.33950424
223.73067141]]
复制代码
作者:
xsoft
时间:
2020-2-3 15:37
谢谢老师提供的资料。
欢迎光临 东方耀AI技术分享 (http://www.ai111.vip/)
Powered by Discuz! X3.4