MOTIP2: Spatial Priors for End-to-End Multi-Object Tracking
B. Roussel, D. Bouet, L. Chen, P. Perrault
British Machine Vision Conference (BMVC)2026ConferenceAccepted, to appear
CVBenoît Roussel
Download PDF ↓Current position
PhD student, computer vision
2023–Idemia & Centrale Lyon·Courbevoie & Lyon
Thesis: End-to-End Multi-Object Tracker for Handling Occlusions. Detection and association trained jointly in a single model, with an eye on long-range tracking through occlusions and on the cost of the labelled video such models demand.
- Advisors
- Liming Chen (LIRIS, École Centrale de Lyon) and Pierre Perrault (Idemia).
- Teaching
- Lectures, lab sessions and talks on tracking and deep learning at Télécom Paris, Sorbonne (UPMC) and Centrale Lyon.
- Supervision
- An M2 intern on real-time end-to-end multi-object tracking.
Prior experience
Semi-professional e-sport player (sabbatical)
2023League of Legends·France
Six months competing in France’s third division league and in LAN events, with a stint in the second division; reached the top 300 players in Europe.
Research intern, computational imaging
2022EPFL, LCAV·Lausanne
Reconstruction for lensless cameras using a generative prior. A generator pre-trained on FFHQ raised PSNR by 0.8 to 9.8 dB depending on the mask; mask creation in the lab’s libraries went from 5.57 s to 35.1 ms.
Research intern, computer vision
2021Dassault Aviation·Saint-Cloud
Automatic aircraft anchoring, from three minutes by hand to thirty seconds. Camera calibration, dataset creation, and deployment to HoloLens through WinML and ONNX.
Research intern, machine learning
2020ONERA, The French Aerospace Lab·Palaiseau
A self-supervised depth estimation technique used to refine the output of a model-based depth map.
Education
Master 2 MVA (mathematics, vision, learning)
2021–2022ENS Paris-Saclay·Gif-sur-Yvette
Graduated with high honors. Applied mathematics for data, image and signal: convex optimization, computational statistics, object recognition, 3D computer vision, point clouds, topological data analysis, graphs in machine learning.
- Tutoring
- Undergraduate electrical engineering at Institut Villebon-Georges Charpak (Orsay): practical experiments and engineering projects, in a format adapted to a cohort recruited for potential rather than prior results.
ARIA, research year in artificial intelligence
2020–2021ENS Paris-Saclay·Gif-sur-Yvette
A research project at Centre Borelli on motor styles in healthy and pathological populations, alongside advanced MVA coursework, then six months at Dassault Aviation.
Formation Saphire, then Master 1 E3A (André Ampère track)
2018–2020ENS Paris-Saclay·Cachan
Two years, both graduated with high honors: a multidisciplinary engineering year, whose year-long project was the fastest autonomous car the school’s Saph Team Racing competition would allow, then a master’s in electrical engineering with added coursework in AI: control systems, signal and image processing, machine learning for time series, video and images.
CPGE, physics and technology
2016–2018Lycée Jules Ferry·Versailles
Two-year intensive preparation for the competitive entrance examinations to the French Grandes Écoles.
Publications
Multi-Object Tracking as Identity Correction
B. Roussel, K. Steunou, D. Martin, P. Perrault
HAL preprint2026Preprint
Skills
- Research
- Multi-object tracking, detection and segmentation, 3D scene understanding, computational imaging
- Programming
- Python, PyTorch, PyTorch3D, JAX, C/C++, Matlab, SQL
- Tooling
- Distributed training (HPC, Slurm), GCP, Linux, shell, Git, LaTeX
- Languages
- French (native), English (full professional, C1 Advanced), German (limited working)