What is the default loss function used in the U-Net implementation in MATLAB for semantic segmentation tasks?

Hi,

Can you please provide A DOCUMENT that explicitly confirms what the default loss function used in the U-Net implementation in MATLAB is for image semantic segmentation tasks? In other words, I need a document that confirms the default loss function used in the U-Net implementation in MATLAB for image semantic segmentation tasks.

Thank you!

Answer 1

[Pratham Shah]

If you are using the unetLayers function, default loss function will be "Cross-Entropy". You can check that on the documentation of pixelClassificationLayer.

[Gobert]

Thank you @Pratham Shah

Can you share a document/link that explicitly shows what the mathematical formula of 'crossentropyex' is?

[Pratham Shah]

You can check the formula on crossentropy. Hope this helps :)

[Gobert]

If Loss function for training, specified as 'crossentropyex', stands for Cross Entropy Function for k Mutually Exclusive Classes ( https://se.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.classificationoutputlayer.html?searchHighlight=crossentropyex&s_tid=srchtitle_crossentropyex_3 ) the formula corresponding to 'crossentropyex'would be the following?