Semantic Segmentation on Driving Datasets

Here I have implemented Segmentation on Cityscapes (Lower Resolution Dataset), and Camvid Dataset using U-Net Model with cross entropy loss for 10-15 epochs.

Even for low number of training run model seems to give good result on test dataset.

Using more sopishticate loss function like Dice loss can definitely increase accuracy.