Join a leading research team at Inria Paris for a fully-funded 3-year PhD position focused on integrating physical understanding into Vision Foundation Models (VFM). This interdisciplinary research bridges machine learning, computer vision, and physics, and is funded by PR[AI]RIE-PSAI.
🎓 PhD Scholarship Opportunity at Inria Paris: Physics-Grounded Vision Foundation Models
🧑🎓 Designation
PhD Candidate – Full-time, 3 years
📋 Table: Scholarship Overview
| Field | Details |
|---|---|
| Title | Physics-Grounded Vision Foundation Models |
| Location | Inria Paris, France |
| Duration | 3 years |
| Funding Body | PR[AI]RIE-PSAI |
| Supervisors | Raoul de Charette (Inria), Tuan-Hung Vu (Inria / Valeo.ai) |
| Start Date | September – November 2025 |
| Application Deadline | May 20, 2025 |
| Interview Dates | May 21–27, 2025 |
| Acceptance Notification | Conditional by May 30, Final by mid-June 2025 |
🧠 Research Area
- Computer Vision
- Vision Foundation Models (VFM)
- Physics-informed Machine Learning
- Generative AI
- Visual Reasoning and Dynamics
📍 Location
Inria Paris – Astra Project Team
A renowned research institute located in Paris, working on sustainable mobility, safety, and AI for visual scene understanding.
Website: https://astra-vision.github.io
✅ Eligibility / Qualification
- Master’s degree in a relevant field (e.g., Computer Science, Physics, AI, Engineering)
- Scientific excellence (publications are a plus)
- Knowledge of foundation models and machine learning
- Strong coding proficiency
- Commitment to diversity, openness, and academic excellence
📝 Description
The goal of the PhD is to build physics-aware Vision Foundation Models (VFM) that go beyond semantic understanding and can model Newtonian dynamics such as gravity, force, motion, and collisions. Research will proceed in three phases:
- Evaluation of existing VFM’s physics understanding using novel pixel-level annotated datasets.
- Design and training of physics-grounded VFMs through supervised and self-supervised learning.
- Exploration of extensions to generative models, non-rigid dynamics (fluids, deformables), and enhanced training strategies.
Scientific dissemination is expected through top-tier conferences such as CVPR, NeurIPS, and ICLR. Open-source contributions and international collaboration are encouraged.
📬 How to Apply
Send the following documents by email to raoul.de-charette@inria.fr:
- CV / Resume
- Names and contact information of two referees
- Motivation letter (1 page max)
- Transcripts and diplomas
⏰ Last Date to Apply
Tuesday, May 20, 2025
