Point clouds are essential in various fields like autonomous driving and robotics, demanding efficient downsampling methods. Traditional approaches lack adaptability to different task requirements. REPS introduces a reconstruction-based scoring strategy that evaluates point importance through reconstruction processes, ensuring preservation of structural features. The Global-Local Fusion Attention module integrates local and global features for high-quality reconstruction and sampling effects. Experimental results demonstrate superior performance across tasks.
Sang ngôn ngữ khác
từ nội dung nguồn
arxiv.org
Thông tin chi tiết chính được chắt lọc từ
by Guoqing Zhan... lúc arxiv.org 03-11-2024
https://arxiv.org/pdf/2403.05047.pdfYêu cầu sâu hơn