Result exception

[APotatoBoy]

Thanks for your sharing! I have tried your released code, the acc results I got are as follows: cub_baseline: test_top1: 85.92% cub_s3n:test_branch1_top1: 87.40% | test_branch2_top1: 86.54% | test_branch3_top1: 86.67% | test_branch4_top1: 84.16%

Not as excellent as was posted in your paper, 88.5%

[Yao-DD]

Could you please share your operation environment？

[xmczh003]

the acc results I got are as follows: test_branch1_top1: 87.83% | test_branch2_top1: 87.21% | test_branch3_top1: 86.62% | test_branch4_top1: 85.74%

[Yao-DD]

One of the most important requirements is using 2 Tesla P100. Other kinds of GPU, such as V100 or K80 will lead to slightly lower accuracy.

[zlweiyang]

Thanks for your sharing! I have tried your released code, the acc results I got are as follows: cub_baseline: test_top1: 85.92% cub_s3n:test_branch1_top1: 87.40% | test_branch2_top1: 86.54% | test_branch3_top1: 86.67% | test_branch4_top1: 84.16%

Not as excellent as was posted in your paper, 88.5%

what's the meanings of test_branch1_top1,test_branch2_top1,test_branch3_top1,test_branch4_top1？

[gengyanlei]

Excuse me, which one of test_branch1_top1, test_branch2_top1, test_branch3_top1, test_branch4_top1 is the most suitable evaluation index?

[Yao-DD]

test_branch1_top1