Skip to content
Benoît Roussel

Benoît Roussel

Hello! I am a PhD student in a CIFRE program between Idemia Public Security and Centrale Lyon, in the LIRIS lab. I work on multi-object tracking: following multiple objects through a video while keeping track of their identities, a task that is simple to describe but hard to solve. My research explores how systems can learn to track by jointly learning detection and association, with the goal of building accurate, efficient trackers suited to real-world deployment.

Beyond my thesis, I’m driven by a desire to understand how things work and not just how to use them. I enjoy digging into the underlying principles and building the intuition needed to develop new ideas. I was drawn to AI because I wanted to explore problems across different fields and turn that understanding into something useful.

News

  1. MOTIP2 is out on arXiv and HAL, ahead of the BMVC proceedings.

  2. MOTIP2 accepted at BMVC 2026: spatial priors for end-to-end multi-object tracking.

  3. MOTIC, multi-object tracking as identity correction, is out as a preprint on HAL.

Selected publications