Expression editability

[tyrink]

During model training and testing, does the fine-grained description of facial local regions include expression information from the reference face? And this seems to weaken the expression editability of the model.