Frédéric Fortier-Chouinard

I am a PhD student in Electrical Engineering at Université Laval, where I have the chance to do research under the supervision of Prof. Jean-François Lalonde. I collaborate with Depix Technologies and receive funding from an FRQNT scholarship.

I work on adding physics-based controls to diffusion models, in particular, for the task of relighting. I also worked on virtual object insertion using diffusion models and neural radiance fields.

I obtained my Bachelor's in Computer Engineering at Université Laval. I pursued an exchange semester in 2022 at EPFL, Switzerland. During my undergrad, I completed three internships where I worked on high-performance computing (HPC) software at Calcul Québec, integrated a machine-learning pipeline in the cloud at Dimonoff, and researched GAN-based high-resolution inpainting methods with Prof. Lalonde.

CV  /  Email  /  GitHub  /  Google Scholar  /  LinkedIn

Research

I'm interested in computer vision, machine learning, graphics, and image generation models.

SpotLight: Shadow-Guided Object Relighting via Diffusion

SpotLight allows precise local lighting control by specifying the desired shadows of the inserted object. This approach accurately reshades the object and properly harmonizes the object with the target background, without any training.

ZeroComp: Zero-shot Object Compositing from Image Intrinsics via Diffusion

Zero-shot 3D object compositing approach that does not require paired composite-scene images during training and can be easily extended to 2D object compositing and material editing.

PanDORA: Casual HDR Radiance Acquisition for Indoor Scenes

PanDORA is a system that allows for HDR capture of indoor scenes using a simple capture device, composed of two cameras. The captured frames are fed two a NeRF-based algorithm that reconstructs the scene’s high dynamic range, allowing applications such as virtual object relighting.

Other Projects

CinematicSpark.com

CinematicSpark is an easy-to-use web application for creating personnalized images of a subject (e.g. a person) using only a handful of low-quality photos as input. It brings cutting-edge techniques in personnalized image generation in the hands of beginner users.

MC Hub

Open source web application for managing cloud high-performance computing (HPC) clusters that I developed during my internship at Calcul Québec. MC Hub is still maintained and used for training HPC users.

Design and source code from Jon Barron's and Leonid Keselman's websites.