Tuning small semantic segmentation models via knowledge distillation

The demand for accurate and efficient driver assistance systems is increasing with the rapid development of the autonomous driving industry. Semantic segmentation models play a critical role in enabling the safe and reliable operation of self-driving vehicles. However, the computational requirements of semantic segmentation models pose challenges for real-time deployment on resource-constrained devices. Developers are thus compelled to utilize less complex model architectures, generally achieving inferior results. This thesis aims to enhance the accuracy of a lightweight semantic segmentation model designed for autonomous vehicle applications, through the application of knowledge distillation.