Is the comparison fair?

[hopef]

Thanks for your excellent work.

I have some doubts.
Based on the comparison, the biggest differences between SRSTE and Bi-Mask are the following:

1. SRSTE: FirstConv runs on SparsityConv and uses "nn.CrossEntropy" in the training phase.
2. Bi-Mask: FirstConv runs on DenseConv and uses "LabelSmooth(0.1)" in the training phase.

We know that these two differences can often lead to significant accuracy gaps.
Looking forward to your reply.

Thanks