About the performance

[WangWenhao0716]

Thanks for the interesting work you have done. However, I have some questions about performance. In your paper, you say the performance of MEB-Net is 76.0%, which is much better than MMT. But as you pointed out, it is DenseNet-121 rather than ResNet-50 that obtains this performance. And if you use ResNet-50 as the backbone to test, the performance is similar to MMT (Fig 4). By the way, I have tried an experiment which is similar to yours: I use three ResNet-50 in MMT rather than two, however, there is no significant improvement. As a result, I think the improved performance coming from a stronger baseline rather than your training strategy. To prove your opinion, it suitable to compare your results with MMT (DenseNet-121 BackBone) and show your superiority. Thanks again and look forward to your reply.