Elias Ramzi - AI Research scientist

I am a Research Scientist at valeo.ai, working on deep learning for autonomous driving. My main focus is end-to-end driving: learning to plan directly from sensor data. Around it I work on world models that predict how a scene will unfold, on self-play in simulation, and on vision-language models for reasoning and explainability. Recent projects include Pictura, a simulator for training driving policies by self-play in the perspective view, and VaViM & VaVAM, a video world model and its action model. I also co-supervise two PhD students, one on LLMs/VLMs and one on world models and reinforcement learning.

Before joining valeo.ai, I earned a PhD in computer vision at Cnam, supervised by Nicolas Thome (Sorbonne Université), Nicolas Audebert (IGN) and Clément Rambour (Cnam), with Xavier Bitot (Coexya) as industrial advisor. My thesis — awarded the AFRIF Prix de Thèse — focused on ranking-loss optimization and hierarchical learning for image retrieval (ROADMAP, HAPPIER, SupRank).

News

======

• Pictura is accepted at an ECCV 2026 workshop: a GPU-accelerated simulator that renders every agent’s egocentric view, and the first large-scale driving self-play policy trained directly from perspective images; [paper] [project page] [code].
• TOAD is online: test-time trajectory optimization that improves existing end-to-end planners without retraining; [paper].
• Franca, on scalable visual representation learning, is accepted at CVPR 2026; [paper].
• DRIV-EX, on counterfactual explanations for driving LLMs, is accepted at ACL Findings 2026, congrats Amaia; [paper].
• My thesis was awarded the Prix de Thèse by AFRIF 🎉
• We have released a tech report and fully open-sourced code and weights for VaViM & VaVAM. This project builds a world model composed of a next frame predictor (VaVIM) and an action model (VaVAM); [paper] [code].
• LLM-wrapper, which allows black-box fine-tuning of VLMs, has been accepted at ICLR 2025, congrats Amaia; [paper] [code].
• SupRank is accepted at TPAMI, it is the first of its kind hierarchical landmark retrieval dataset; [paper] [code] [dataset].

Publications

======

Pictura: Perspective-View Self-Play at Scale for Driving

Yuan Yin, Elias Ramzi, Marc Lafon, Valentin Charraut, Victor Bares, Yihong Xu, Éloi Zablocki, Alexandre Boulch, Thibault Buhet, Andrei Bursuc, et al.: Pictura: Perspective-View Self-Play at Scale for Driving. ECCV Workshop (2026).

Test-Time Trajectory Optimization for Autonomous Driving

Yihong Xu, Éloi Zablocki, Yuan Yin, Elias Ramzi, Ellington Kirby, Alexandre Boulch, Matthieu Cord: Test-Time Trajectory Optimization for Autonomous Driving. arXiv preprint (2026).

DRIV-EX: Counterfactual Explanations for Driving LLMs

Amaia Cardiel, Éloi Zablocki, Elias Ramzi, Eric Gaussier: DRIV-EX: Counterfactual Explanations for Driving LLMs. ACL findings (2026).

Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning

Shashanka Venkataramanan, Valentinos Pariza, Mohammadreza Salehi, Lukas Knobel, Spyros Gidaris, Elias Ramzi, Andrei Bursuc, Yuki M. Asano: Franca: Nested Matryoshka Clustering for Scalable Visual Representation Learning. CVPR (2026).

LLM-wrapper: Black-Box Semantic-Aware Adaptation of Vision-Language Models for Referring Expression Comprehension

Amaia Cardiel, Éloi Zablocki, Elias Ramzi, Oriane Siméoni, Matthieu Cord: LLM-wrapper: Black-Box Semantic-Aware Adaptation of Vision-Language Models for Referring Expression Comprehension. Internation Conference on Learning Representations (ICLR 2025).

VaViM and VaVAM: Autonomous Driving through Video Generative Modeling

Florent Bartoccioni, Elias Ramzi, Victor Besnier, Shashanka Venkataramanan, Tuan-Hung Vu, Yihong Xu, Loick Chambon, Spyros Gidaris, Serkan Odabas, David Hurych, Renaud Marlet, Alexandre Boulch, Mickael Chen, Éloi Zablocki, Andrei Bursuc, Eduardo Valle, Matthieu Cord: VaViM and VaVAM: Autonomous Driving through Video Generative Modeling. ArXiv Preprint 2025.

Optimization of Rank Losses for Image Retrieval

Elias Ramzi, Nicolas Audebert, Clément Rambour, André Araujo, Xavier Bitot, Nicolas Thome: Optimization of Rank Losses for Image Retrieval. In IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI, 2025).

GalLoP: Learning Global and Local Prompts for Vision-Language Models

Marc Lafon, Elias Ramzi, Clément Rambour, Nicolas Audebert, Nicolas Thome: GalLoP: Learning Global and Local Prompts for Vision-Language Models. European Conference on Computer Vision (ECCV 2024).

ITEM: Improving Training and Evaluation of Message-Passing based GNNs for top-k recommendation

Yannis Karmim, Elias Ramzi, Raphaël Fournier-S 'Niehotta, Nicolas Thome: ITEM: Improving Training and Evaluation of Message-Passing based GNNs for top-k recommendation. TMLR (2024).

Hybrid Energy Based Model in the Feature Space for Out-of-Distribution Detection

Marc Lafon, Elias Ramzi, Clément Rambour, Nicolas Thome: Hybrid Energy Based Model in the Feature Space for Out-of-Distribution Detection. International Conference on Machine Learning (ICML 2023).

Hierarchical Average Precision Training for Pertinent Image Retrieval

Elias Ramzi, Nicolas Audebert, Nicolas Thome, Clément Rambour, Xavier Bitot: RHierarchical Average Precision Training for Pertinent Image Retrieval. In: European Conference on Computer Vision. Springer (ECCV, 2022).

Robust and Decomposable Average Precision for Image Retrieval

Elias Ramzi, Nicolas Thome, Clément Rambour, Nicolas Audebert, Xavier Bitot: Robust and decomposable average precision for image retrieval. Advances in Neural Information Processing Systems 34 (NeurIPS, 2021).