LucidFusion: Reconstructing 3D Gaussians with Arbitrary Unposed Images

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: He, Hao, Liang, Yixun, Wang, Luozhou, Cai, Yuanhao, Xu, Xinli, Guo, Hao-Xiang, Wen, Xiang, Chen, Yingcong
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929748179943424
author He, Hao
Liang, Yixun
Wang, Luozhou
Cai, Yuanhao
Xu, Xinli
Guo, Hao-Xiang
Wen, Xiang
Chen, Yingcong
author_facet He, Hao
Liang, Yixun
Wang, Luozhou
Cai, Yuanhao
Xu, Xinli
Guo, Hao-Xiang
Wen, Xiang
Chen, Yingcong
contents Recent large reconstruction models have made notable progress in generating high-quality 3D objects from single images. However, current reconstruction methods often rely on explicit camera pose estimation or fixed viewpoints, restricting their flexibility and practical applicability. We reformulate 3D reconstruction as image-to-image translation and introduce the Relative Coordinate Map (RCM), which aligns multiple unposed images to a main view without pose estimation. While RCM simplifies the process, its lack of global 3D supervision can yield noisy outputs. To address this, we propose Relative Coordinate Gaussians (RCG) as an extension to RCM, which treats each pixel's coordinates as a Gaussian center and employs differentiable rasterization for consistent geometry and pose recovery. Our LucidFusion framework handles an arbitrary number of unposed inputs, producing robust 3D reconstructions within seconds and paving the way for more flexible, pose-free 3D pipelines.
format Preprint
id arxiv_https___arxiv_org_abs_2410_15636
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle LucidFusion: Reconstructing 3D Gaussians with Arbitrary Unposed Images
He, Hao
Liang, Yixun
Wang, Luozhou
Cai, Yuanhao
Xu, Xinli
Guo, Hao-Xiang
Wen, Xiang
Chen, Yingcong
Computer Vision and Pattern Recognition
Recent large reconstruction models have made notable progress in generating high-quality 3D objects from single images. However, current reconstruction methods often rely on explicit camera pose estimation or fixed viewpoints, restricting their flexibility and practical applicability. We reformulate 3D reconstruction as image-to-image translation and introduce the Relative Coordinate Map (RCM), which aligns multiple unposed images to a main view without pose estimation. While RCM simplifies the process, its lack of global 3D supervision can yield noisy outputs. To address this, we propose Relative Coordinate Gaussians (RCG) as an extension to RCM, which treats each pixel's coordinates as a Gaussian center and employs differentiable rasterization for consistent geometry and pose recovery. Our LucidFusion framework handles an arbitrary number of unposed inputs, producing robust 3D reconstructions within seconds and paving the way for more flexible, pose-free 3D pipelines.
title LucidFusion: Reconstructing 3D Gaussians with Arbitrary Unposed Images
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.15636