segmentation models - make num_classes and softmax non must

[yonigottesman]

Currently all the segmentation models deep_lab_v3, segformer etc.. create the final layer with num_classes and always does softmax. Maybe we should have an activation parameter and if its None dont do any activation.
Some classification models require sigmoid and sometimes I want the unet for generative tasks without and activation

[sachinprasadhs]

Thanks for reporting the issue.
We have the re enhanced DeepLab and other models in Keras Hub which has a option to provide the activation .
For more details check https://github.com/keras-team/keras-hub/blob/master/keras_hub/src/models/deeplab_v3/deeplab_v3_segmenter.py