Multi-Object Tracking and Person Re-Identification

This lecture covers Multi-Object Tracking (MOT) and Person Re-Identification (ReID). It starts with an overview of MOT, discussing the challenges and related tasks. The lecture then delves into the problem statement of MOT, distinguishing between online and offline tracking methods. It explains different tracking paradigms, focusing on Tracking-by-Detection as the most popular approach. The lecture also presents a simple online tracking-by-detection method, detailing the process of detection, association, and the Hungarian algorithm. Additionally, it explores the concept of Person Re-Identification, explaining the retrieval task and the metric learning aspect. The lecture concludes by discussing the state-of-the-art methods for online tracking-by-detection and the training process of a ReID model.