RA-YOLOX: Re-parameterization align decoupled head and novel label assignment scheme based on YOLOX.

•We propose a lightweight RepA decoupled head that aligns two tasks with minimal increase in parameters compared to the YOLOX head. Additionally, we apply a structural re-parameterization method to the detection head, leading to improved detector performance.•We propose a novel label assignment which divides positive and negative samples more effectively and designs loss weight function more precisely.•We provide three different sizes of lite models, RA-YOLX-S, RA-YOLOX-tiny, and RA-YOLOX-nano, which outperform YOLOX with the same size by average precision of 2.3%, 1.5% and 1.7% respectively on the MS COCO-2017 validation set.