In neural network training have many type of loss function. In this project is mainly on segmentation. And usually CNN (Convolutional neural network) is a pixel wise prediction. that mean it predict by each pixel probability. So i find several loss function information for what i need.
1.BCEloss (Binary Cross Entropy )
BCEloss is for Binary labels training, It is the basic loss for binary pixel wise prediction.
2.CrossEntropyLoss
It can training for multi labels And it is applied logSoftmax and NLLLoss. It is the common loss function for multi labels segmentation.
3. Dice loss
It is the loss function to calculate the distance between prediction and true mask like the below image.
Comments