why I got the negative loss ?

[zzdang]

I got the nagetive loss use the unet ,I don't know what's the wrong in my code. I use the gt (maks) of 2 class,background and foregroud,with the mask pixels are 255 .The mask png is 1 channel. I use image_util.ImageDataProvider to provide data,unet.Unet is the network. Thank you for helping me

[PhySci]

Do you use "class_weights" in the model definition?

[jakeret]

Depending on the loss you are using this actually valid

[PhySci]

I see that cross-entropy loss function gives negative value for my model. But according to the definition, it should be strictly positive. Any ideas? I'm trying to sort out the problem and realize what's going on.

[jakeret]

I see, that's every odd. Haven't seen that before