HZQ-GitHub
HZQ-GitHub
If I use Windows, how to do ?
@wadesunyang I have the same issues with you about how to ignore other "hardly useful "samples for landmark task? Do you solver it now?
@zfc929 Hi, I notice your answer that use label 0 instead of -1. Yeah, I try it, But,the effect is very poor. All 5 landmarks concentrate on the ONE point...
@zfc929 Hi, Thanks for your reply. I want to know that have you rewiten the eculidean_loss_layer in caffe to support multi-task training and how to rewrite? Have you ever tried...