Dear Tong,
First of all, thank you for this page, it has proven quite useful. While figuring out how the code works, I think I have stumbled upon a small documentation mistake. In line 71 and 72 of optimal_transport.py, x1 is described as Tensor, shape (bs, *dim) represents the source minibatch. I think it should be the target minibatch.
Thank you!
Best regards