About Me
Hey there! I'm Victor — a Research Scientist focused on Computer Vision. Since 2018, when I first explored GANs during an internship, I’ve been passionate about image and video generation. Over time, I’ve worked with diffusion models, flow matching, and recently explored discrete approaches inspired by GPT-style and BERT-style, the latter enabling fast parallel token sampling for image generation. My work has been published at top venues like CVPR, ICCV, and ICLR. Beyond my research, I’m interested in building real-time models for autonomous driving and creating efficient retrieval systems to navigate large datasets of unlabeled images and videos.
Publications
Halton Scheduler For Masked Generative Image Transformer
ICLR 2025
Authors: V. Besnier, M. Chen, D. Hurych, E. Valle, M. Cord
A novel scheduling strategy for MaskGIT, using with Halton sequences. Improves quality and diversity.
Read on arXiv →
Don't Drop Your Samples! Coherence-aware training for Conditional Diffusion
CVPR 2024
Authors: N. Dufour, V. Besnier, V. Kalogeiton, D. Picard
We introduce coherence-aware training of conditional diffusion models, reducing the need to filter dataset.
Read on arXiv →
Triggering Failures: OOD Detection in Semantic Segmentation
ICCV 2021
Authors: V. Besnier, A. Bursuc, D. Picard, A. Briot
Explores learning from local adversarial attacks for more robust out-of-distribution detection in segmentation models.
Read on arXiv →
VaViM and VaVAM: Autonomous Driving through Video Generative Modeling
ArXiv 2025
Authors: F. Bartoccioni, E. Ramzi, V. Besnier, S. Venkataramanan et al.
We explore the potential of large-scale generative video models for autonomous driving, introducing an open-source auto-regressive video model (VaViM) and its companion video-action model (VaVAM).
Read on arXiv →
Supervised Anomaly Detection for Complex Industrial Images
CVPR 2024
Authors: A. Baitieva, D. Hurych, V. Besnier, O. Bernard
A supervised anomaly detection method tailored to high-complexity industrial imagery settings.
Read on arXiv →
Learning Uncertainty For Safety-Oriented Semantic Segmentation
ICIP 2021
Authors: V. Besnier, D. Picard, A. Briot
We present a framework for modeling predictive uncertainty in segmentation tasks related to autonomous driving.
Read on arXiv →
This dataset does not exist: training models from generated images
ICASSP 2020
Authors: V. Besnier, H. Jain, A. Bursuc, M. Cord, P. Perez
Investigating the use of synthetic datasets for training deep models, with a focus on image quality and task generalization.
Read on arXiv →📰 News
Jan 2025
I became a Valeo Expert — 6 exciting years so far! 🎉
Jun 2024
Presented 'Don't drop your sample!' and 'SegAD' at CVPR 2024 in Seattle. 🚀
Feb 2023
Started at Valeo.ai Prague as a Research Scientist in video & image synthesis.
Nov 2022
🎓 Got my PhD in ML for autonomous systems!
Positions
2023–Present: Research Scientist, Valeo.ai
Focus: Image and video synthesis with Diffusion Models, MaskGIT, Auto-regressive models, and GANs.
Collaborators: Patrick Pérez, Matthieu Cord
Location: Prague, Czech Republic
2019–2022: PhD Candidate, Valeo
Thesis: Safety of automotive systems through Machine Learning
Advisors: David Picard, Alexandre Briot
Location: Créteil, France
2019: Research Intern, Valeo.ai
Topic: Generative Adversarial Networks and latent space optimization for image classification
Advisors: Matthieu Cord, Himalaya Jain
Location: Paris, France
2018: Research Intern, LIP6
Topic: Unsupervised super-resolution using Generative Adversarial Networks
Advisor: Ludovic Denoyer
Location: Paris, France
Education
2019–2022: PhD, École des Ponts ParisTech (ENPC)
Thesis: Safety of autonomous systems using Machine Learning
Location: Champs-sur-Marne, France
2017–2019: Master's Degree, Sorbonne Université (Paris VI)
Program: DAC – Data Science and Machine Learning
Distinction: Graduated with honors
Location: Paris, France
2014–2017: Bachelor's Degree, UPMC (Paris VI)
Major: Computer Science and Mathematics
Distinction: Graduated with honors
Location: Paris, France
Contact
Feel free to reach out or connect with me: