End-to-end object detection using a query-selection encoder with hierarchical feature-aware attention

• The proposed query-selection encoder (QSE) significantly accelerates training convergence and improves detection accuracy for end-to-end object detectors. • The hierarchical feature-aware attention (HFA) mechanism suppresses similar feature representations and highlights discriminative ones, expediting feature selection. • QSE is versatile and can be seamlessly integrated into both CNN- and Transformer-based detection architectures. • Extensive experiments on MS COCO, CrowdHuman, and PASCAL VOC demonstrate that QSE enhances end-to-end performance with fewer training epochs.