Flow Distillation Sampling: Regularizing 3D Gaussians with Pre-trained Matching Priors

Lin-Zhuo Chen Nanjing University, Kangjie Liu Nanjing UniversityYoutian LinNanjing UniversitySiyu ZhuFudan UniversityZhihao LiHuawei Noah’s Ark LabXun CaoNanjing UniversityYao Yao Nanjing University · 2025

Read Flow Distillation Sampling: Regularizing 3D Gaussians with Pre-trained Matching Priors by Lin-Zhuo Chen Nanjing University, Kangjie Liu Nanjing UniversityYoutian LinNanjing UniversitySiyu ZhuFudan UniversityZhihao LiHuawei Noah’s Ark LabXun CaoNanjing UniversityYao Yao Nanjing University (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-2502-07615.txt.

Description

Lin-Zhuo Chen and co-authors add learned matching knowledge to 3D scene reconstruction, in this 2025 paper. A popular method builds scenes from soft 3D blobs and renders them fast, but its geometry sags where cameras never looked. Extra views must be invented to teach it better shape.

Their technique samples new viewpoints near the real ones. A pre-trained matching model supplies a flow field for those views, which is compared with the flow implied by the current geometry. The mismatch steers the blobs toward truer shapes. Tests on depth, mesh building, and fresh view rendering showed clear gains over leading methods.

Catalog record

CREATORLin-Zhuo Chen Nanjing University, Kangjie Liu Nanjing UniversityYoutian LinNanjing UniversitySiyu ZhuFudan UniversityZhihao LiHuawei Noah’s Ark LabXun CaoNanjing UniversityYao Yao Nanjing University
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-07615

Browse the full library · About the Tower of Babel