Hello, in your paper, you present the results on the VisDA-2017 dataset. However, in the code, there is no implementation for testing on the VisDA-2017.

The difference loss in 《Domain Separation Networks》 is the Frobenius norm of a matrix composed of vectors, where the elements of the matrix are the dot products of these vectors. A smaller value indicates lower similarity between the vectors, which I understand. However, in your paper, you mention calculating the Frobenius norm of a matrix obtained by multiplying two vectors, where the matrix elements are products of the vector components. This seems unrelated to vector similarity. What is the significance of this?