Abstract: Single-pixel imaging (SPI) is a sensing technique that uses a single-pixel detector to acquire spatially resolved measurements of a scene. Compressive SPI reduces acquisition time by measuring only a fraction of the Hadamard patterns, with the sampling strategy strongly impacting the reconstruction quality. In this paper, we study SPI subsampling strategies, including popular approaches from the literature, and introduce a new strategy, called seeded intermediate subsampling, that achieves state-of-the-art performance at low sampling rates. Our experimental results on an SPI testbed validate these findings. Note: Date estimée (numéro 34(20)) ; DOI vérifié via HAL. Abstract à vérifier contre la version finale Optics Express.