A Novel Dynamic Model Selection Approach for Image Object Detection
Ye Liu, Keyu Wu, Zhong Liu · Electronics Letters · 2025
ABSTRACT Model selection is crucial for image object detection, enabling the selection of the best model for varying tasks. This paper introduces a novel dynamic model selection approach that adapts to underlying scenarios. The problem is formulated as a joint clustering and model assignment task, where clustering reveals the data's inherent structure, and the best model is assigned to each cluster. To solve this, we propose a spatial structure‐preserving genetic algorithm, an efficient optimization method that integrates spatial information into genetic operations, ensuring faster and more stable convergence. Experimental results show that our approach outperforms individual models, highlighting the effectiveness and efficiency of dynamic model selection.