Éloi Zablocki

Research scientist at valeo.ai

photo_eloi_zablocki.jpg

I am a research scientist at valeo.ai, working on:

  • Vision and language, Explainability
  • Autonomous driving, dynamic scene understanding and forecasting

Before joining Valeo, I completed a Ph.D. at Sorbonne Université in 2019 on multi-modal machine learning with language and vision, supervised by Patrick Gallinari, Laure Soulier, and Benjamin Piwowarski. Prior to this, I earned an MSc from ENS Paris-Saclay (“Master MVA”) and École Polytechnique (X2012).

News

Feb 2025 New release from the team: VaViM & VaVAM. VaViM is a 1.2B parameter video generative model trained on 1,800+ hours of raw Youtube videos. Its derived video-action model, VaVAM, achieves state-of-the-art results on NeuroNCAP driving benchmark. We open-source code, model weights, training recipes and scaling laws.
Feb 2025 New preprint: GaussRender. A plug-and-play 3D-to-2D reprojection loss using Gaussian splatting enhances 3D semantic occupancy learning from multiple cameras.
Jan 2025 Annealed Winner-Takes-All for Motion Forecasting is accepted at ICRA 2025. We show that using an annealing loss enhances training stability and performance of state-of-the-art trajectory prediction models.
Jan 2025 LLM-wrapper is accepted at ICLR 2025. It shows that LLMs can learn to adapt black-box VLMs for new tasks and domains, by wrapping and reasoning on the vision models’ outputs.
Dec 2024 New preprint: PPT. Pre-training trajectory prediction models with pseudo-labeled trajectories improves performance, efficiency, and generalization.
Nov 2024 New preprint :gift:: GIFT. A framework for generating global, interpretable textual explanations of vision classifiers, combining counterfactual visual explanations with VLMs and LLMs.
Sep 2024 I participate to ECCV 2024 in Milan. I will present UniTraj, LLM-wrapper, ReGentS, and Valeo4Cast.
Aug 2024 3 papers accepted in workshops at ECCV: LLM-wrapper, ReGentS, and Valeo4Cast.
Jul 2024 UniTraj is accepted at ECCV 2024.
Jun 2024 Valeo4Cast wins the Argoverse 2 “Unified Detection, Tracking and Forecasting” challenge held at CVPR WAD 2024.
Mar 2024 PointBeV is accepted at CVPR 2024.
Jan 2024 Our paper on Motion Forecasting with Real-World Perception Inputs is accepted at ICRA 2024.

Current Students

Alumni

Scientific Service

Reviewer:
Conferences: CVPR 2021–2025, ECCV/ICCV 2021–2024, ICLR 2025, ICML 2025, WACV 2024, IROS 2022, AAAI 2021
Journals: IJCV 2022,2025, TPAMI 2021–2022, T-ITS 2023
Workshops: ROAD++ (@ICCV 2023)

PhD Jury Member:
Florent Bartoccioni (Invited Member, 2023)