所属单位：计算机科学与工程学院

发表刊物：2023 4th International Conference on Computer Vision, Image and Deep Learning (CVIDL)

项目来源：自选课题

关键字：real-time; semantic segmentation; autonomous driving

摘要：One of the fundamental tasks in computer vision is semantic segmentation, which is increasingly being widely applied, for example in autonomous driving. However, past research has often resulted in large computational overhead and high latency. To address this issue, we propose a lightweight architecture called Pyramid Simple Detail Injection Transformer (PSDFormer). Experimental results demonstrate the superiority of our approach in lightweight semantic segmentation. On the Cityscapes dataset, PSDFormer achieves an mIoU of 66.78 while achieving 57.1 fps.

合写作者：陈晨,鲍学良,仲昭昊

第一作者：刘冰

论文类型：论文集

页面范围：2

是否译文：否

发表时间：2023-05-14