GSemSplat: Generalizable Semantic 3D Gaussian Splatting from Uncalibrated Image Pairs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Xingrui, Lan, Cuiling, Zhu, Hanxin, Chen, Zhibo, Lu, Yan
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913622898245632
author Wang, Xingrui
Lan, Cuiling
Zhu, Hanxin
Chen, Zhibo
Lu, Yan
author_facet Wang, Xingrui
Lan, Cuiling
Zhu, Hanxin
Chen, Zhibo
Lu, Yan
contents Modeling and understanding the 3D world is crucial for various applications, from augmented reality to robotic navigation. Recent advancements based on 3D Gaussian Splatting have integrated semantic information from multi-view images into Gaussian primitives. However, these methods typically require costly per-scene optimization from dense calibrated images, limiting their practicality. In this paper, we consider the new task of generalizable 3D semantic field modeling from sparse, uncalibrated image pairs. Building upon the Splatt3R architecture, we introduce GSemSplat, a framework that learns open-vocabulary semantic representations linked to 3D Gaussians without the need for per-scene optimization, dense image collections or calibration. To ensure effective and reliable learning of semantic features in 3D space, we employ a dual-feature approach that leverages both region-specific and context-aware semantic features as supervision in the 2D space. This allows us to capitalize on their complementary strengths. Experimental results on the ScanNet++ dataset demonstrate the effectiveness and superiority of our approach compared to the traditional scene-specific method. We hope our work will inspire more research into generalizable 3D understanding.
format Preprint
id arxiv_https___arxiv_org_abs_2412_16932
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle GSemSplat: Generalizable Semantic 3D Gaussian Splatting from Uncalibrated Image Pairs
Wang, Xingrui
Lan, Cuiling
Zhu, Hanxin
Chen, Zhibo
Lu, Yan
Computer Vision and Pattern Recognition
Modeling and understanding the 3D world is crucial for various applications, from augmented reality to robotic navigation. Recent advancements based on 3D Gaussian Splatting have integrated semantic information from multi-view images into Gaussian primitives. However, these methods typically require costly per-scene optimization from dense calibrated images, limiting their practicality. In this paper, we consider the new task of generalizable 3D semantic field modeling from sparse, uncalibrated image pairs. Building upon the Splatt3R architecture, we introduce GSemSplat, a framework that learns open-vocabulary semantic representations linked to 3D Gaussians without the need for per-scene optimization, dense image collections or calibration. To ensure effective and reliable learning of semantic features in 3D space, we employ a dual-feature approach that leverages both region-specific and context-aware semantic features as supervision in the 2D space. This allows us to capitalize on their complementary strengths. Experimental results on the ScanNet++ dataset demonstrate the effectiveness and superiority of our approach compared to the traditional scene-specific method. We hope our work will inspire more research into generalizable 3D understanding.
title GSemSplat: Generalizable Semantic 3D Gaussian Splatting from Uncalibrated Image Pairs
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2412.16932