I'm trying to build a regression model in keras, the input will be a normalized greyscale image, and the output of the model will be a normalized greyscale image too.
I'm aware that a sigmoid at the output is not ideal for regression models since it gives a probability between [0 , 1] which is good for classification problems. but since my data is already normalized between [0 , 1], would a sigmoid output be a good idea ?
Thank you
Generally the Linear activation function is used for regression problems.
Sigmoid activation function is not used for normalizing the data. This is mostly used for binary classification problems when there are 2 values(0 or 1) that need to be predicted for each binary class. Values below 0.5 will be assumed as class 0 and >0.5 will be taken as class 1(for the class between 0,1).
If you have a multi classification problem, you may need to use Softmax activation function to get the probability of each class and then use argmax() to get the high probability index to define the specific class among all other probabilities.
You can follow this link for more details.
