Question about torch.svd

[XMU-qcj]

svd_cpu: the updating process of SBDSDC did not converge (error: 2)

[paul007pl]

I also met this issue frequently for clean data registration.
I think it should be a bug with the svd function, because many previous methods also observe this issue.

Based on my experience, here are two suggestions:

1. Use another random seed, and re-train.
2. Using the hyper-parameter estimation sub-network may help.