LLM‑Guided Taxonomy and Hierarchical Uncertainty for 3D Point Cloud Active Learning

Chenxi Li, Nuo Chen, Fengyun Tan, Yantong Chen, Bochun Yuan, Tianrui Li, Chongshou Li Southwest Jiaotong University, Chengdu, China{lcx041511, chennuo2022, tfy, 2022116015, ybc23793}@my.swjtu.edu.cn{trli, lics}@swjtu.edu.cn · 2025

Read LLM‑Guided Taxonomy and Hierarchical Uncertainty for 3D Point Cloud Active Learning by Chenxi Li, Nuo Chen, Fengyun Tan, Yantong Chen, Bochun Yuan, Tianrui Li, Chongshou Li Southwest Jiaotong University, Chengdu, China{lcx041511, chennuo2022, tfy, 2022116015, ybc23793}@my.swjtu.edu.cn{trli, lics}@swjtu.edu.cn (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-18924.txt.

Description

Chenxi Li and co-authors use chat models to build label trees that guide which points of a 3D scan to hand tag, in a paper from 2025. Hand tags for 3D scenes cost dear, so teams tag a few points and let a model learn the rest. Past picks treat tags as flat, which wastes the rank that links broad kinds to fine ones.

The chat model drafts a rank tree of labels on its own. A spread rule then pours doubt scores down and up that tree, so picks stay spread in space and rich in kind. On two room scan sets the method lifts scores by up to four points at tag shares as low as a few in ten thousand. The authors cast rank doubt as a strong new frame.

Catalog record

CREATORChenxi Li, Nuo Chen, Fengyun Tan, Yantong Chen, Bochun Yuan, Tianrui Li, Chongshou Li Southwest Jiaotong University, Chengdu, China{lcx041511, chennuo2022, tfy, 2022116015, ybc23793}@my.swjtu.edu.cn{trli, lics}@swjtu.edu.cn
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-18924

Browse the full library · About the Tower of Babel