Road Detection and Segmentation Using FCN8 having VGG16 as encoder and trained on KITTI dataset.
The model is trained for 35 epochs and it was able to detect the road with an accuracy of 93%. If we can train the model for about 100 epochs or more, we can achieve more accurate results.
you can download the best_model weights after 35 epochs here
Download the KITTI Road segmentation dataset used to train the model here
Sample Predictions: