Scholarlink

Labeling point cloudsPoint clouds completely is extremely time-consuming and expensive. As larger point cloud datasets with billions of points become increasingly common, we wonder whether full annotation is truly necessary. Our results show that models designed under the assumption of complete supervisionComplete supervision only suffer a slight degradation when using 1% random point annotationsPoint annotations.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Semantic Segmentation of Point Clouds

  • 章节

  • 发表日期：

    2026年

查看全文
  • Yulan Guo,
  • Sheng Ao,
  • Zhiheng Fu,
  • Hao Liu,
  • Qingyong Hu

摘要

Labeling point cloudsPoint clouds completely is extremely time-consuming and expensive. As larger point cloud datasets with billions of points become increasingly common, we wonder whether full annotation is truly necessary. Our results show that models designed under the assumption of complete supervisionComplete supervision only suffer a slight degradation when using 1% random point annotationsPoint annotations.