Initialization Of Decoder Layer

[Parikshat1998]

In the paper, you mentioned that you are initializing the last layer of the decoder using the PCAs of Vertex displacements in the training data. So, these vertex displacements are computed due to expression changes in the training meshes or they are generic and capture the overall vertex displacements of the meshes?

The name of the file init_expression_basis.npy, so why it is named as expression basis, and also if I explore the shape of the tensor it is (5023*3,100). Can you please provide more idea on this initialization technique?