issues about Channel-wise Correlation

[shuaiyongjiang]

In Channel-wise Correlation Module(line 97):

  exemplar_att = exemplar_att.view(-1, self.channel, fea_size[0], fea_size[1])  # N,C1,H*W -> N,C1,H,W
  exemplar_out = self.conv1(exemplar_att + exemplar)

  query_att = query_att.view(-1, self.channel, fea_size[0], fea_size[1])  # N,C2,H*W -> N,C2,H,W
  query_out = self.conv1(query_att + query)

two "self.conv1" are sharing weight?

[MathLee]

Thanks for pointing out this issue. This is our mistake.
"query_out = self.conv1(query_att + query)" should be "query_out = self.conv2(query_att + query)".

[MathLee]

This mistake is an engineering code error and will not affect our motivation to design the network. Thanks again!