Tuan-Hung Vu
- Senior Research Scientist · valeo.ai
- Part-time Researcher, Astra-vision · Inria Paris
- Assistant Professor · VinUniversity
I am a Senior Research Scientist at valeo.ai (R&D Technical Manager, Valeo Expert) and a part-time researcher in the computer vision group Astra-vision at Inria Paris. I am also an Assistant Professor at VinUniversity. I received my PhD from École Normale Supérieure under the supervision of Ivan Laptev. Before that, I obtained an engineering degree from Télécom Paris and, in parallel, a “Master 2” degree in Mathematics, Machine Learning and Computer Vision (MVA) from École Normale Supérieure Paris-Saclay in 2014.
My research focuses on deep learning for scene understanding, with an emphasis on making perception models robust to distribution shift: domain adaptation and generalization, vision-language models, 3D perception, and generative models for autonomous driving.
News
- Vanilla ViT for automotive point cloud segmentation is accepted to WACV 2027.
- Our video diffusion scaling-law paper receives the Runner-up Best Paper Award (oral) at the ECCV 2026 DriveX Workshop.
- New preprint on scaling laws of video diffusion models trained on 5,500 hours of driving.
- NewtPhys preprint: do foundation models understand Newtonian physics?
- Vanilla ViT for automotive point cloud segmentation preprint is online.
- OccAny on generalized urban 3D occupancy is accepted to CVPR 2026. Project page
- Driving on Registers preprint is online (accepted to CVPR 2026).
- Our paper on multimodal distillation for 3D segmentation under domain shift is presented at BMVC 2025.
Show older news
- IPA, an input-projection framework for efficient foundation model adaptation, is online (accepted to TMLR).
- FLOSS on training-free open-vocabulary segmentation is online (ICCV 2025). Project page
- VaViM and VaVAM: autonomous driving through video generative modeling. Code
- The ProLIP preprint on few-shot learning with CLIP is online. Code
- Results of the BRAVO semantic segmentation challenge at UNCV, ECCV 2024.
- The GenVal paper on synthetic evaluation in semantic segmentation is accepted to ECCV 2024. Code
- The TTYD paper on robust source-free domain adaptation is accepted to ECCV 2024. Code
- Valeo4Cast wins the Argoverse 2 “Unified Detection, Tracking, and Forecasting” challenge.
- The FAMix paper on domain generalized semantic segmentation is accepted to CVPR 2024.
- The SALUDA paper on point-cloud domain adaptation is accepted to 3DV 2024. Code
- The PØDA paper on prompt-driven zero-shot domain adaptation is accepted to ICCV 2023. Code
- Submission is open for the BRAVO workshop. Deadline is 19/07/2023.
- “BRAVO: Robustness and Reliability of Autonomous Vehicles in the Open-world” is accepted as an ICCV 2023 workshop.
- Our work SALUDA on point-cloud domain adaptation is online. Inference code is released.
- Our work PØDA on prompt-driven zero-shot domain adaptation is online. Code is released.
- Top reviewer, NeurIPS 2022.
- The code of DenseMTL is released.
- One paper accepted to WACV 2023.
- Our work on “Cross-task Attention Mechanism for Dense Multi-task Learning” is online. Code is released.
- We’re organizing a workshop on Weakly Supervised Computer Vision at the Deep Learning Indaba 2022.
- Two papers, “CSG0: Continual Urban Scene Generation with Zero Forgetting” and “Multi-Head Distillation for Continual Unsupervised Domain Adaptation in Semantic Segmentation”, are accepted to the CVPR 2022 CLVision Workshop.
- Our paper “Cross-modal Learning for Domain Adaptation in 3D Semantic Segmentation” is accepted to T-PAMI.
- The CSG0 preprint “Continual Urban Scene Generation with Zero Forgetting” is online.
- Our paper on boundless unsupervised domain adaptation is accepted to CVIU.
- One paper on multi-target domain adaptation is accepted to ICCV 2021.
- The Semantic Palette code is released.
- The preprint and video of Semantic Palette are available.
- Our work on confidence estimation and its use in improving domain adaptation is accepted to T-PAMI. Paper
- Outstanding reviewer, CVPR 2021.
- Outstanding reviewer, ICLR 2021.
- Our paper Semantic Palette on layout-scene generation is accepted to CVPR 2021.
- The xMUDA journal preprint “Cross-modal Learning for Domain Adaptation in 3D Semantic Segmentation” is online.
- The ConDA preprint “Confidence Estimation via Auxiliary Models” is online.
- The xMUDA code is released.
- Our paper ESL on entropy-guided pseudo-labels for UDA is accepted to the CVPR 2020 Workshop on Scalability in Autonomous Driving.
- The BUDA preprint on boundless unsupervised domain adaptation is online.
- Our paper xMUDA on cross 2D-3D unsupervised domain adaptation is accepted to CVPR 2020.
- Demo at CES, Las Vegas, on pedestrian behavior understanding and forecasting. Press: 1, 2, 3.
- The DADA code is released.
- The zero-shot semantic segmentation code is released.
- The xMUDA preprint on cross 2D-3D unsupervised domain adaptation is online. Demo
- Our paper Zero-shot Semantic Segmentation is accepted to NeurIPS 2019.
- Invited talk at the BNP Paribas AI Summer School.
- Our paper DADA is accepted to ICCV 2019.
- The ADVENT code is released.
- The ZS3 preprint on zero-shot semantic segmentation is online.
- Keynote talk at ULAD, the 1st Workshop on Unsupervised Learning for Automated Driving, at IV 2019.
- The DADA preprint on unsupervised domain adaptation is online.
- Our paper ADVENT is accepted to CVPR 2019 as an oral presentation.
- The Tube-CNN preprint on object detection in videos is online.
- The ADVENT preprint on unsupervised domain adaptation is online.
- I successfully defended my PhD!
- Our MemNet paper on object detection in videos is accepted to WACV 2019.
- I joined valeo.ai as a research scientist.
- Summer internship at NEC Labs, Cupertino.
- Code for the ICCV 2015 paper is released.
- A paper on human head detection in movies is accepted to ICCV 2015.
- Dataset for the ECCV 2014 paper is released.
- A paper on action prediction is accepted to ECCV 2014.
Publications
Full and up-to-date list on Google Scholar.
2026
2025
2024
2023
2022
2021
2020
2019
Academic Service
Organization
- BRAVO Semantic Segmentation Challenge @ UNCV, ECCV 2024
- BRAVO Workshop: Robustness and Reliability of Autonomous Vehicles in the Open-world, ICCV 2023
- Workshop on Weakly Supervised Computer Vision, Deep Learning Indaba 2022
Reviewing
- Conferences
- CVPR, ICCV, ECCV (2018–) · ICLR, NeurIPS (2021–) · AAAI · ICRA · IROS
- Journals
- IJCV · T-PAMI · CVIU
Education
- PhD in Computer Vision, École Normale Supérieure — Advisor: Ivan Laptev
- Master 2 MVA — Mathematics, Machine Learning & Computer Vision, École Normale Supérieure Paris-Saclay
- Engineering degree, Télécom Paris