Open oadams opened 4 years ago
In this implementation as well as others I've found online, there are separate weights for each layer for the feature fusion:
https://github.com/toandaominh1997/EfficientDet.Pytorch/blob/fbe56e58c9a2749520303d2d380427e5f01305ba/models/bifpn.py#L147
In the paper it appears as though the same weights are used between layers (see the equations on page 4 column 2).
It seems it was a deliberate choice to use more parameters here. Or am I missing something in the paper?
In this implementation as well as others I've found online, there are separate weights for each layer for the feature fusion:
https://github.com/toandaominh1997/EfficientDet.Pytorch/blob/fbe56e58c9a2749520303d2d380427e5f01305ba/models/bifpn.py#L147
In the paper it appears as though the same weights are used between layers (see the equations on page 4 column 2).
It seems it was a deliberate choice to use more parameters here. Or am I missing something in the paper?