OP2GS: Object-Aware 3D Gaussian Primitives via Dual Opacity

Guiyu Liu, Niklas Vaara, Janne Mustaniemi, Juho Kannala, Janne Heikkilä · 2026

Read OP2GS: Object-Aware 3D Gaussian Primitives via Dual Opacity by Guiyu Liu, Niklas Vaara, Janne Mustaniemi, Juho Kannala, Janne Heikkilä (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-2605-20044.txt.

Description

Guiyu Liu and four co-workers give 3D Gaussian splatting primitives a clear object identity with a system called OP2GS. Each primitive gains an instance identity plus a second opacity for object-mask rendering, while the original opacity still handles visual reconstruction. That dual-opacity split lets mislabelled Gaussians stay visible in images yet turn transparent in the object-mask branch. Object-level semantic descriptors, gathered by multi-view aggregation, then avoid storing features per Gaussian, and tests show strong open-vocabulary performance at much lower overhead.

Catalog record

CREATORGuiyu Liu, Niklas Vaara, Janne Mustaniemi, Juho Kannala, Janne Heikkilä
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE44240
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.20044

Browse the full library · About the Tower of Babel