Hello, I failed to reproduce the reported results on metr-la dataset. Thus, I came to your implementation. However, the results turned to be worse than the reported one as well.
I'm really curious about this thing. Thanks for your implementation anyway.