How to use SigmoidCrossEntropyLoss in caffe for binary class classification?

Question

How to use SigmoidCrossEntropyLoss in caffe for binary class problem? Can you give me a prototxt example?

Shai · Accepted Answer · 2016-04-23T22:00:08.413

2

You need your label input to be either zero or one per training sample. other than that

 layer {
   name: "loss"
   type: "SigmoidCrossEntropyLoss"
   bottom: "predict1d"
   bottom: "label"
   top: "loss"
 }

edited Apr 23 '16 at 22:00

answered Apr 22 '16 at 14:17

Shai

I tried the same way. Before this layer I had a layer with number of output is equals to 1. I also tried "SoftmaxWithLoss". It seems "SigmoidCrossEntropyLoss" is very much slower than "SoftmaxWithLoss". Do you have any idea about why this happens? – user570593 Apr 22 '16 at 14:21
@user570593 unfortunately I have no experience with this loss layer – Shai Apr 23 '16 at 18:39

1 Answers1