For example, in PointNet, you see the 1D convolutions with the following channels 64 -> 128 -> 1024. Why not e.g. 64 -> 1024 -> 1024 or 1024 -> 1024 -> 1024?
Asked
Active
Viewed 128 times
1