Long-read sequencing technologies have great potential for the comprehensive discovery of structural variation (SV). However, the accurate genotype assignment for SV is still a challenge since the unavoidable factors like specific sequencing errors or limited coverages. Herein, we propose cuteSV2, a fast and accurate long-read-based re-genotyping approach that is able to force-calling genotypes for the given records. cuteSV2 is an upgraded version of cuteSV and applies an improved strategy of the refinement and purification of the heuristic extracted signatures through spatial and allele similarity estimation. The benchmarking results on several baseline evaluations demonstrate that cuteSV2 outperforms the-state-of-art methods and has stable and robust capability in practice.