Scientific Achievements

Applied research

I have contributed to applied research projects deployed as demos or in production, including (but not limited to):

re-identifying soccer players through jersey number recognition;
real-time ball tracking in low-resolution sports videos;
predicting the primary topic of webpages from limited, noisy annotations (built from scratch and deployed in production for the 2018 World Cup);
detecting and reducing the influence of traffic spikes on webpage view counts (deployed within the forecast tool);
understanding daily human activities in home environments (Mobile Mii);
inferring disparity maps around individuals’ faces.

Publications

My research has been focused on representation learning for video understanding when annotated data is limited, and on adapting vision models under distribution shift. A complete list is on Google Scholar.

2025

MicroFlow: Domain-Specific Optical Flow for Ground Deformation Estimation in Seismic Events

Juliette Bertrand, Sophie Giffard-Roisin, James Hollingsworth, Julien Mairal

Preprint

project page · code

2023

Test-time Training for Matching-based Video Object Segmentation

Juliette Bertrand*, Giorgos Kordopatis-Zilos*, Yannis Kalantidis, Giorgos Tolias

NeurIPS

project page · code

2023

Rethinking Matching-Based Few-Shot Action Recognition