我用了AI Studio上分享的螺丝检测的YOLOv3代码来训练检测帽子的项目,但是经过120个epoch后模型都保存好了,但是预测没结果也没报错,我用的paddlepaddle1.4与螺丝帽检测的1.3不同,是什么原因啊?
{'0.0': 'hat'}
predict cost time:0.05 sec
(1, 18, 8, 8) [48, 50, 81, 100, 205, 191]
(1, 18, 16, 16) [6, 8, 13, 15, 22, 34]
result boxes: []
result scores: []
result labels: []
疑似问题部分:
出错的好像是这一段,在公开的YOLOv3检测螺丝的项目中这一段的运行结果如下,PaddlePaddle版本为1.3:
[[ 97.11353 17.630293 137.32475 68.692726]
[136.23756 63.942276 175.13948 121.69328 ]
[ 90.670334 89.6736 125.56294 156.17635 ]
[133.46649 73.625626 175.05615 137.1838 ]
[137.66988 16.807865 162.5731 51.553066]
[173.74017 43.098305 199.53033 77.52401 ]
[115.501 58.09054 143.95004 92.091705]
[ 81.76508 91.59957 106.37437 125.18125 ]
[186.76233 84.68567 212.8393 119.21594 ]
[172.15247 138.71683 197.75769 177.75513 ]
[ 91.45226 151.2745 117.3847 184.89523 ]
[173.36816 142.33243 197.16626 178.54166 ]
[155.71378 163.03499 180.57947 196.35558 ]]
[0.9982137 0.9701177 0.98191935 0.9551219 0.99935347 0.99913776
0.99924386 0.999869 0.99988425 0.9994111 0.9997894 0.5891546
0.9999442 ]
[0. 0. 0. 0. 1. 1. 1. 1. 1. 1. 1. 1. 1.]
但在我的自建项目中,PaddlePaddle版本为1.4,输出全为空:
[]
[]
[]
这就百思不得其解了,是因为版本问题吗?如果是哪又应该怎么改呢?