In this project, we use Fully Convolutional Network (FCN) to classify every pixel in an image to be road or not road. It is a basic step in scene understanding. This project implements the idea from this paper: Fully Convolutional Networks for Semantic Segmentation
Make sure you have the following is installed:
Download the Kitti Road dataset from here. Extract the dataset in the data folder. This will create the folder data_road with all the training and test images.
- number of epochs: 50
- batch size: 16
- keep_prob: 0.75
- learning_rate: 0.0001
Run the following command to run the project:
python main.py
The loss is decreasing over training and converge at the end. The loss value vs. training time is shown below.
Several sample images with labeled pixels are shown below.
