New publication at BMVC 2026!
I am proud to announce the publication of our new paper, “TriCCOT: Tri-part Convolutional Conformal Transformer for Onboard Space Object Detection”, at such a top-tier conference in computer vision.
The preprint is already available there: https://arxiv.org/abs/2609.08659
This work aims to reduce the domain shift gap encountered when deploying AI models onboard satellites, where models must operate directly on raw, noisy imagery under strong computational constraints.
To address this, we introduce TriCCOT, combining a convolutional Region Proposal Network, conformal prediction to improve proposal coverage, and Aper-GATES, a hardware-aware transformer designed for FPGA accelerators.
The architecture was validated on real raw images, showing strong robustness to spatial blur and signal-dependent noise, and was successfully deployed on a Xilinx Versal FPGA without modifying the underlying accelerator architecture.
A big thanks to my co-authors Marjorie Bellizzi, Julia Cohen, PhD, Stephane May, as well as the Centre National d’Études Spatiales and IRT Saint Exupéry for supporting this work.