From Handcrafted to End-to-End Learning and Back: A Journey Through Multi-Object Tracking Paradigms

Explore multi-object tracking (MOT) paradigms in this lecture from the AI Doctoral Academy's Excellence Series, where Prof. Dr. Laura Leal-Taixé examines the evolution from tracking-by-detection to end-to-end learning approaches and back again. Delve into the complexities of simultaneous track initialization, identity management, and spatiotemporal trajectory analysis while comparing traditional methods with modern tracking-by-regression and tracking-by-attention paradigms. Discover why the field has come full circle to reconsider tracking-by-detection as a potentially state-of-the-art solution for contemporary MOT challenges.