SegviGen: Repurposing 3D Generative Model for Part Segmentation

Haoran Feng, Zehuan Huang, Haohua Chen, Wenbo Nie, Shaohua Hou, Keqing Fan Pan Hu Sheng Wang Buyu Li Lu Sheng, Beihang University Renmin University of China Tsinghua University OriginArk Project page: https://fenghora.github.io/SegviGen-Page/ · 2026

Read SegviGen: Repurposing 3D Generative Model for Part Segmentation by Haoran Feng, Zehuan Huang, Haohua Chen, Wenbo Nie, Shaohua Hou, Keqing Fan Pan Hu Sheng Wang Buyu Li Lu Sheng, Beihang University Renmin University of China Tsinghua University OriginArk Project page: https://fenghora.github.io/SegviGen-Page/ (2026) 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-2603-16869.txt.

Description

Li and a large team introduce SegviGen, which turns a 3D generative model into a part segmentation tool. Older routes lift 2D priors into 3D and blur borders, or need huge labeled datasets. SegviGen instead uses the priors inside a pretrained generative model. It encodes a 3D asset and predicts colors that mark parts on a matched reconstruction. One framework covers interactive, full, and 2D-guided segmentation. The authors report gains of 40 percent on interactive and 15 percent on full segmentation, using only 0.32 percent of the labeled training data.

Catalog record

CREATORHaoran Feng, Zehuan Huang, Haohua Chen, Wenbo Nie, Shaohua Hou, Keqing Fan Pan Hu Sheng Wang Buyu Li Lu Sheng, Beihang University Renmin University of China Tsinghua University OriginArk Project page: https://fenghora.github.io/SegviGen-Page/
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31921
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.16869

Browse the full library · About the Tower of Babel