Sequential spacecraft pose estimation via visual geometry grounded transformers and learnable token merging

doi:10.1016/j.ast.2026.112310

Sequential spacecraft pose estimation via visual geometry grounded transformers and learnable token merging

2026-04-08

0

PRE

AI

DOI：10.1016/j.ast.2026.112310

Original

Original request for help

Save

Abstract

En 中文

• Learnable token merging reduces 13% computation with only 0.5% accuracy drop than no merging. • Achieves 57.0% ADD-0.1d accuracy on SwissCube, outperforming prior methods by 21.6%. • Operates at 68 GFLOPs, enabling real-time edge deployment.

Keywords:

token merging

visual geometry

transformers

pose estimation

edge deployment

Sequential spacecraft pose estimation via visual geometry grounded transformers and learnable token merging

Abstract

Journal

Organization