TGRNet icon indicating copy to clipboard operation
TGRNet copied to clipboard

环境同步后,Lloc各项结果还是接近于0

Open zdmwang opened this issue 1 year ago • 3 comments

          @EmperorKaiser 谢谢。

@hsdjkfnsfc 抱歉回复比较晚。我在cuda10.2下复现了你的问题,但是切换到cuda9.2后,结果正常。所以,很可能是不同cuda版本下,软件的影响,特别时torch geometric。 我的测试环境:

  • cuda 9.2
  • torch 1.7.0+torchvision 0.8.0
  • torch-cluster 1.5.9
  • torch-geometric 1.6.3
  • torch-scatter 2.0.6
  • torch-sparse 0.6.9
  • torch-spline-conv 1.2.1

Originally posted by @xuewenyuan in https://github.com/xuewenyuan/TGRNet/issues/2#issuecomment-972423748

zdmwang avatar Mar 31 '23 06:03 zdmwang

环境同步后,Lloc各项指标还是接近于0,测试环境为:

  • torch 1.7.0+cu92
  • torchvision 0.8.1+cu92
  • torch-cluster 1.5.9
  • torch-geometric 1.6.3
  • torch-model-archiver 0.7.1
  • torch-scatter 2.0.6
  • torch-sparse 0.6.9
  • torch-spline-conv 1.2.1 test_tablegraph-24k最终结果是: IoU: 0.500, Sloc, Recall: 0.9224, Precision: 0.9086, Hmean(F1): 0.9155 IoU: 0.500, Lloc, Acc_all: 0.0000, Acc_rowSt: 0.0099, Acc_rowEd: 0.0140, Acc_colSt: 0.0199, Acc_colEd: 0.0047 IoU: 0.500, F(beta=0.5): 0.000

zdmwang avatar Mar 31 '23 06:03 zdmwang

环境同步后,Lloc各项指标还是接近于0,测试环境为:

  • torch 1.7.0+cu92
  • torchvision 0.8.1+cu92
  • torch-cluster 1.5.9
  • torch-geometric 1.6.3
  • torch-model-archiver 0.7.1
  • torch-scatter 2.0.6
  • torch-sparse 0.6.9
  • torch-spline-conv 1.2.1 test_tablegraph-24k最终结果是: IoU: 0.500, Sloc, Recall: 0.9224, Precision: 0.9086, Hmean(F1): 0.9155 IoU: 0.500, Lloc, Acc_all: 0.0000, Acc_rowSt: 0.0099, Acc_rowEd: 0.0140, Acc_colSt: 0.0199, Acc_colEd: 0.0047 IoU: 0.500, F(beta=0.5): 0.000

zdmwang avatar Mar 31 '23 06:03 zdmwang

@zdmwang 仅提供这些信息,我很难帮助到你,建议你check中间结果,或者重新训练一个模型

xuewenyuan avatar Apr 03 '23 09:04 xuewenyuan