Gaspard Beaudouin

I am currently a Research Scientist Intern at Mistral AI on the Multimodal team, working on Vision Encoder and multimodal models. Upon completion, I will earn a Master’s degree in Mathematics, Vision and Learning (MVA) from ENS Paris-Saclay and graduate from École nationale des ponts et chaussées, where I specialized in Mathematics and Computer Science.

My research interests focus on generative modelling, World Models, flow matching, Representation Learning (JEPA, SSL for videos ..), VLMs and reasoning. I conducted research on flow matching at Harvard AI and Robotics Lab, supervised by Prof. Mengyu Wang. I have also worked on Large Language Models during an internship at ChapsVision, more precisely about training for agentic/tool calling abilities (SFT and DPO).

I am the first author of Delta Rectified Flow Sampling for Text-to-Image Editing, accepted at CVPR 2026. I also co-authored SplitFlow: Flow Decomposition for Inversion-Free Text-to-Image Editing, accepted at NeurIPS 2025. Thanks to Minghan Li, Sung-Hoon Yoon, Jaeyon Kim and Mengyu Wang for their incredible collaboration. I also co-authored PAGE-4D: Disentangled Pose and Geometry Estimation for 4D Perception, accepted at ICLR 2026, with Kaichen Zhou.

You can find my code and projects on GitHub.