A precise and real-time abdominal multi-organ segmentation method is of crucial importance for its practical application. In this study, we use a two-phase strategy to address this issue. In the phase one, we quickly localize the abdominal region, while the second phase focuses on fine segmentation of this region. This work builds upon last year’s efforts. To improve inference efficiency, we designed a Lightweight Attention-based Convolutional Block for the phase two and incorporated it into the decoder. Additionally, the preprocessing process has been further optimized. The results on leaderboard validated promising performance, achieving an average score of 90.02% and 95.51% for the DSC and NSD. Additionally, the method’s average running time on public validation is 16.34 s in our laptop. In summary, this strategy effectively ensures the possibility of achieving high precision with low latency. Our code is available at: https://github.com/JCXiong1227/FLARE2024 .