Open husthkk opened 3 years ago
Just one 3x3 Conv layer. The output channel is determined by the size of the kernel you want. To reduce the amount of parameters, we share the kernel weight among channels. So the output channel is set 9 for 3x3 kernel. For position-wise convolution operation, we recommend referring to the code of CARAFE.
In the paper,you said that the Gc represents the dynamic filter generator and Kc are the learned example-wise kernel weights. So can you explain the specific structure of Gc? Thank you.