SoN: Selective Optimal Network for smartphone-based indoor localization in real-time

Kyungsu Lee, Haeyun Lee, Jae Youn Hwang

Abstract: Highlights•A novel framework, Selective Optimal Network (henceforth SoN), is proposed for high-speed classification applications.•An effective path selection algorithm is developed for the SoN to help it select an optimal network for input samples according to their complexity in the classification.•Additionally, classification-complexity, quantifying the level of classification-complexity for an input image, is proposed to select the optimal backbone architecture from ones of different depths.•The semi-supervised methodology to search the optimal classification-complexity corresponding to the individual images is devised in order to fine-tune the optimal level of difficulty while training SoN.•Generally, the SoN for the smartphone-based IPS provides faster prediction time (78.59%) and better accuracy (1.50%) than other SotA models.

External IDs:doi:10.1016/j.eswa.2025.126639