数据标注方法及系统
By performing voxelization and sparse interaction on LiDAR point clouds and 3D images, and combining large language models and visual models, the consistency and efficiency issues of 3D data annotation were solved, achieving high-precision 3D bounding box determination and 2D segmentation mask generation, thus improving the flexibility and efficiency of annotation.
CN121545158BActive Publication Date: 2026-07-17WUHAN UNIV OF TECH
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WUHAN UNIV OF TECH
- Filing Date
- 2025-12-19
- Publication Date
- 2026-07-17
Smart Images
Figure CN121545158B_ABST
Abstract
本发明提供了一种数据标注方法及系统，其方法包括：分别对目标场景的激光雷达点云和三维图像进行体素化处理，得到多尺度的体素特征；对同一尺度的点云体素特征与图像体素特征进行稀疏交互，生成多模态稀疏交互体素特征；确定目标场景中至少一个目标对象的三维边界框及对应的置信度；根据用户提供的用于描述目标对象的自然语言指令，通过大语言模型生成自然语言指令对应的描述文本；基于描述文本驱动预设的提示式视觉模型，通过迭代交互方式，生成目标对象对应的二维分割掩码；将三维边界框与二维分割掩码进行匹配，并将二维分割掩码所表征的语义类别标注至所匹配的三维边界框上，减少了人工干预频率，提升了标注效率、灵活性和一致性。
Need to check novelty before this filing date? Find Prior Art