semantic image segmentation is to label each pixel of an image with a corresponding class
first we use small dataset . Download from this link https://drive.google.com/file/d/0B0d9ZiqAgFkiOHR1NTJhWVJMNEU/view?usp=sharing
input image shape is (w, h, 3) and the idea is to make output shape is (w, h, n_classes) where n_classes is number of class in image number of diff objects (Max_number of Classes)