No ReLU Mask Code in your MAT Module.

[8414sys]

You have written about the utility of ReLU Mask in your paper. The content was briefly that ReLU Mask, which does not require learning parameters, performs better than DynaST's learnable MLP. The implementation of that part is in models/networks/dynast_transformer.py. I don't think your code has changed at all compared to the original code in DynaST, is that correct? If correct, should I simply apply the ReLU function to the output instead of the corresponding code?