sg-s opened this issue 4 years ago · 0 comments
we need to consider the timestep. Different datasets have differnet resolutions.
The best thing to do is to normalize the data and train on that.