Hello, may I ask why the generator uses binary classification cross entropy? I think the mean square error should be used, because the saliency prediction is not a binary classification problem, but a regression problem.
@SoulCoderCN At the first, they considered MSE as a content loss. Then authors presented their reasons at their paper's 4.th page. Also you can check this slide to get information.
ErolCitak
commented
5 years ago
Hello, may I ask why the generator uses binary classification cross entropy? I think the mean square error should be used, because the saliency prediction is not a binary classification problem, but a regression problem.