Open Internship Proposals

We are looking for students finishing their MSc with a solid background in computer vision and machine learning, particularly in deep learning with strong PyTorch coding skills. Interns work on research topics, typically for 6 months (usually spring and summer), resulting for a great part in paper submissions to top-tier conferences. Some trainees go on to do a PhD thesis in the lab. We currently have four exciting internship opportunities for MSc students!

How to Apply

Send an email to the supervisors (one email per application) with the following:

• A cover letter explaining your interest and qualifications for the topic.
• Your CV/resume.
• Transcripts of your grades from last year (and this year, if available).

Available Topics:

Universal 2D-3D Transformer
Keywords: Transformer, Representation learning, Self-supervised learning, Multi-modal
Supervisors: Tuan-Hung Vu, Gilles Puy, Spyros Gidaris
This project aims to develop a novel transformer architecture capable of processing 2D and 3D data simultaneously, probing synergistic multi-modal representations between imagery and LiDAR data.

Learning from One Continuous Long-Range Video Stream
Keywords: Video understanding, Continual learning, Transformer, Self-supervised pretraining
Supervisors: Shashanka Venkataramanan, Andrei Bursuc
This internship involves building a video understanding model inspired by human episodic memory to learn continuously from long-range streams. It includes exploring continual learning, memory integration, and advanced pretraining techniques using real-world video datasets.

Scenario Generation for Robust Autonomous Driving using Diffusion Models
Keywords: Diffusion models, Long-tail distribution, Online mapping, Motion prediction
Supervisors: Yuan Yin, Yihong Xu
This internship explores using diffusion models to generate driving scenarios, focusing on map and trajectory creation. The goal is to develop robust, vector-based maps and diverse vehicle behaviors to enhance motion forecasting and planning.

Object Generation from Range Images
Keywords: Diffusion models, Point clouds, Controllable generative models
Supervisors: Nermin Samet, Victor Besnier
This project focuses on generating LiDAR point cloud objects by leveraging pre-trained diffusion models on range image representations. The goal is to improve the controllability of LiDAR object generation in a computationally efficient way.

Alumni interns and visiting students

Projects led or contributed to by interns