I'm a Masters student in the Probabilistic Vision Group at McGill University under the supervision of Professor Tal Arbel.
You can find my complete CV here, and my personal portfolio here.
In our IEEE RA-L paper, we address the challenge of predicting future 3D human poses from past observations. We introduce an open-source library for human pose forecasting, offering multiple models, standardized evaluation metrics, and support for various datasets. Our methods improve forecasting accuracy by modeling both aleatoric and epistemic uncertainty. Experiments on Human3.6M, AMSS, and 3DPW datasets show up to 25% improvement in short-term forecasting without compromising long-term performance. Our code is available online to facilitate further research.
In our ICCV workshop paper, we explore the reconstruction of 3D human-object interactions from images, including both human and object shape and pose estimation. We introduce a novel autoregressive transformer-based variational autoencoder that learns a robust shape prior from extensive 3D datasets. By leveraging the reconstructed 3D human body, our approach enhances object shape and pose estimation. Experimental results on the BEHAVE dataset demonstrate the effectiveness of our method, achieving a 40.7 cm Chamfer distance and showcasing the benefits of learning a shape prior.
2021, made with in pure Bootstrap 4, inspired by Academic Template for Hugo