I have question about motion module.

[justinday123]

In paper, it says that motion module(temporal attention layer) is only trained in stage two, but in train code, it seems like all model's parameter is trained. What is the code that only trains temporal layer?