Semantic Segmentation

We present a Generative Adversarial Network (GAN) model to perform semantic segmentation in automotive videos. We propose a novel procedure for binarization of the original input images by introducing an intermediate Deep neural network that takes as input, the deep features as computed by a pretrained Convolutional Neural Network (CNN). The proposed method is evaluated on two challenging datasets: KITTI and CamVid, and provides competitive performance with state-of-the-art methods.