Since starting my Master's thesis in 2014, I have been exploring *deep learning* approaches to solve problems related to *3D character animation*. These applications include *motion classification, posing, motion completion, motion generation* and *text-to-motion*.
I have built and trained *convolutional, recurrent, attention* and *prototypical neural networks* in settings such as *supervised* and *semi-supervised learning, auto-encoding* and *variational auto-encoding, generative adversarial models*, and *diffusion*. My most recent focus has been to build state-of-the-art text-to-motion diffusion models, while trying to improve generalization and allowing for control and constraints to enable practical use-cases for video-game development.
In recent years, I have worked extensively with *Python* and libraries such as *PyTorch, PyTorch-Lightning, PyTorch3D, OmegaConf* and *Hydra* on the research side, while participating in integration efforts using *Unity, C#* and occasionally *MotionBuilder*.
Developing deep learning models to solve complex 3D animation problems, such as posing, motion completion, motion generation and others, using attention, convolutional and prototypical networks, diffusion models, control nets, and more.
Developing recurrent neural network-based solutions for automatic generation of character animation transitions and semi-supervised approaches for 3D motion classification.
Probabilistic and learning techniques (INF8225)
Advanced computer graphics (INF8702)
Deep Representation Learning of 3D Human Motion with Recurrent Neural Networks
Supervised by Dr. Christopher Pal
Deep Learning for Animation
Co-Supervised by Dr. Christopher Pal and Dr. Michel Gagnon
Software Engineering
With Unity Labs team for AI and Physics Assisted Character Pose Authoring
2nd place in demos category & 3rd place in limited time category with student team
Outstanding scores during Bachelor's degree
Sports & Outdoors
Canoeing and Kayaking
Camping
Volleyball, Badminton
Scuba diving
Music
Guitar
Bass
Drums
Piano
Games
Board games
Video games
Created 2D flash animations that accumulated a total of more than 65M views.
I am one of the five actors of the LaFAN1 dataset, arguably the one with the best dance moves.
Recently, combat moves were performed by a robot.
