computer vision, systems, code
I am a PhD student in Computer Vision based in Angers, France. My work sits around efficient vision transformers, semantic segmentation, and methods that still matter when measured as end-to-end latency.
work
Efficient deep learning for dense prediction and practical deployment.
Rust, Python, and tools that make experiments easier to trust.
Research notes, engineering logs, and the useful parts of failed attempts.
selected
With Pejman Rasti and David Rousseau. Accepted to CVPR 2026 Findings. MPM is a training-free token aggregation method for semantic segmentation, designed to preserve dense reconstruction while reducing real wall-clock latency.
links
The best public trail is code and research.