The last layer of model

vision
1

Hi,
The output of my model is single channel image, and the last layer is a conv. Do I need to add other layer sunch as BN and ReLU after the last layer?

2

Well it depends on several things, the loss, what do you want to predict etcetera…
If you want to predict RGB values then it can be nice and you remove negative values this way. If you want to predict audio between -1 and 1 that would be impossible using a ReLu.
The last activation may be something different from ReLu depending on your output. Even no activation.