two questions about this improvement

[jiaqinxu]

1: I trained it with GOT10K and got a very low results,Is VID necessary?
2: Random noise is used in training, and pictures are input in running time.If you take out any of these four branches, you'll perform poorly.So can I think of it as a four branch network that performs better？