dcn输入
xiximelon opened this issue · 0 comments
xiximelon commented
以下调用DCN部分的代码,fea = (self.dconv_1(fea, offset1))中self.dconv_1输入的是特征和偏移量,但是fea特征是连接了support帧和参考帧的特征,是否应该只输入supp的特征进行align?
fea = torch.cat([ref, supp], dim=1)
fea =self.cr(fea)
offset1 = self.off2d_1(fea)
fea = (self.dconv_1(fea, offset1))