What's the purpose of these operations? To compute the values in tangent space that anchored at FM?
In the forward process of ComplexConv2Deffangle, the weight norm method of out_rot is weightNormalize1:
out_rot = (torch.sum(out_rot*weightNormalize1(self.weight_matrix_rot2),2)),
yet the weight norm method of out_abs is weightNormalize:
out_abs = torch.exp(torch.sum(out_abs*weightNormalize(self.weight_matrix_rot2,self.drop_prob),2))
Are there any theoretical explanations for the difference between these weight norm methods above?
Hi, thanks for sharing the code! After reading the paper and the code, I have some questions about the complex conv operation:
In the forward process of ComplexConv2Deffgroup, there are these operations:
What's the purpose of these operations? To compute the values in tangent space that anchored at FM?
In the forward process of ComplexConv2Deffangle, the weight norm method of out_rot is weightNormalize1:
out_rot = (torch.sum(out_rot*weightNormalize1(self.weight_matrix_rot2),2))
, yet the weight norm method of out_abs is weightNormalize:out_abs = torch.exp(torch.sum(out_abs*weightNormalize(self.weight_matrix_rot2,self.drop_prob),2))
Are there any theoretical explanations for the difference between these weight norm methods above?