In the paper's Proposal Generation for Novel Classes. In this section, feature fusion subnetworks are mentioned, including multiplication, subtraction, and concatenation. May I ask, how is it reflected in the code? In the code, I only found the depthwise grouped convolutions for the fusion part. Looking forward to your reply.
In the paper's Proposal Generation for Novel Classes. In this section, feature fusion subnetworks are mentioned, including multiplication, subtraction, and concatenation. May I ask, how is it reflected in the code? In the code, I only found the depthwise grouped convolutions for the fusion part. Looking forward to your reply.