The objectives of this project are to develop a convolutional neural network based on the VGG-16 image classifier for performing semantic segmentation and in order to determine the area in an image that is a road, as well as processing a video stream of data in real time. The KITTI dataset is used for training our model and providing all the test images too.
The network uses a pretrained VGG-16 model that gets converted to a fully convolutional network that has two classes. The final fully connected layer is converted to a 1x1 convolution. Additionally to increase performance the network is using skip connections with 1x1 convolutions on VGG layers that are added element wise and upsampled. A kernel initializer and regularizer are also used in each convolution and transpose convolution layers and have proved to increase performance and reduce loss significantly.
After 30 epochs loss per batch is on average around 0.065. Below you can see some sample images from the last run.
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 a test images.
Run the following command to run the project:
python main.py
Note If running this in Jupyter Notebook system messages, such as those regarding test status, may appear in the terminal rather than the notebook.