Why cross entropy loss is used here?

dmlc / gluon-cv

Gluon CV Toolkit

http://gluon-cv.mxnet.io

Apache License 2.0

5.8k stars 1.21k forks source link

Why cross entropy loss is used here? #305

Open WormCoder opened 5 years ago

WormCoder commented 5 years ago

https://github.com/dmlc/gluon-cv/blob/9669d728d17715586765a9f2b1886fc054e1d6ed/gluoncv/loss.py#L224

In yolov3's paper, the author claimed that mse loss was adopted for box regression. And as far as I know cross entropy loss is for classification problems, so why cross entropy loss is used here?

zhreshold commented 5 years ago

See definition of SigmoidBinaryCE: https://mxnet.incubator.apache.org/api/python/gluon/loss.html?highlight=sigmoid#mxnet.gluon.loss.SigmoidBinaryCrossEntropyLoss

tldr; CrossEntropy != multi class softmax cross entropy