PartField : Learning 3D Feature Fields for Part Segmentation and Beyond

Minghua Liu Mikaela Angelina Uy Donglai Xiang Hao Su, Sanja Fidler Nicholas Sharp Jun Gao · 2025

Read PartField : Learning 3D Feature Fields for Part Segmentation and Beyond by Minghua Liu Mikaela Angelina Uy Donglai Xiang Hao Su, Sanja Fidler Nicholas Sharp Jun Gao (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-2504-11451.txt.

Description

Splitting a 3D object into its parts, like legs and tops of a table, helps computers reason about the physical world. Minghua Liu and co-authors present PartField in this 2025 paper, a model that learns a field of part features over any shape in a single fast pass.

Training mixes labeled shapes with automatic image-based part guesses, taught through contrastive learning. Clustering the learned field yields a tree of parts, from whole object down to fine detail. The method proved up to a fifth more accurate and far faster than recent rivals. Because the field stays consistent across shapes, it also supports matching parts between different objects.

Catalog record

CREATORMinghua Liu Mikaela Angelina Uy Donglai Xiang Hao Su, Sanja Fidler Nicholas Sharp Jun Gao
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-11451

Browse the full library · About the Tower of Babel