Question about --flippingtest

[CallMeNumber1]

When I applied --flippingtest, the shape of inputs and targets from validation set change to (batch_size, 2, 3, 448, 448) and (batch_size, 2) respectively. And later the code fail to work, reporting error in the part of accuracy computing, which bother me a lot. May i ask that have you tried --flippingtest in the released code and how to fix it ?

[mboboGO]

Sorry for the confusing, I will check the code soon.

[Rayyoana]

When I applied --flippingtest, I also meet this problem.
And the experiment result reproduced without fine-tuning on the CUB dataset was 48.7%. Which is 7.8% lower than the result of 56.5% in your paper.