Deep Learning for Autonomous Vehicles: Predictive Models

This lecture covers the topic of predictive models and trackers in the context of deep learning for autonomous vehicles. It delves into the challenges of object detection, tracking, and multi-target tracking, discussing various algorithms and strategies. The instructor presents the formulation of single and multi-target tracking problems, along with the mathematical definitions and optimization techniques involved. Additionally, the lecture explores the use of neural networks for tracking, the importance of object representation, and the advancements in learning to track with recurrent neural networks. The session concludes with insights into 3D pedestrian localization, uncertainty estimation, and the application of social cues for tasks like social distancing.