Saved in:
Bibliographic Details
Main Authors: Chen, Zequn, Yang, Jiezhi, Yang, Heng
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.16877
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915033952288768
author Chen, Zequn
Yang, Jiezhi
Yang, Heng
author_facet Chen, Zequn
Yang, Jiezhi
Yang, Heng
contents We present PreF3R, Pose-Free Feed-forward 3D Reconstruction from an image sequence of variable length. Unlike previous approaches, PreF3R removes the need for camera calibration and reconstructs the 3D Gaussian field within a canonical coordinate frame directly from a sequence of unposed images, enabling efficient novel-view rendering. We leverage DUSt3R's ability for pair-wise 3D structure reconstruction, and extend it to sequential multi-view input via a spatial memory network, eliminating the need for optimization-based global alignment. Additionally, PreF3R incorporates a dense Gaussian parameter prediction head, which enables subsequent novel-view synthesis with differentiable rasterization. This allows supervising our model with the combination of photometric loss and pointmap regression loss, enhancing both photorealism and structural accuracy. Given a sequence of ordered images, PreF3R incrementally reconstructs the 3D Gaussian field at 20 FPS, therefore enabling real-time novel-view rendering. Empirical experiments demonstrate that PreF3R is an effective solution for the challenging task of pose-free feed-forward novel-view synthesis, while also exhibiting robust generalization to unseen scenes.
format Preprint
id arxiv_https___arxiv_org_abs_2411_16877
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle PreF3R: Pose-Free Feed-Forward 3D Gaussian Splatting from Variable-length Image Sequence
Chen, Zequn
Yang, Jiezhi
Yang, Heng
Computer Vision and Pattern Recognition
We present PreF3R, Pose-Free Feed-forward 3D Reconstruction from an image sequence of variable length. Unlike previous approaches, PreF3R removes the need for camera calibration and reconstructs the 3D Gaussian field within a canonical coordinate frame directly from a sequence of unposed images, enabling efficient novel-view rendering. We leverage DUSt3R's ability for pair-wise 3D structure reconstruction, and extend it to sequential multi-view input via a spatial memory network, eliminating the need for optimization-based global alignment. Additionally, PreF3R incorporates a dense Gaussian parameter prediction head, which enables subsequent novel-view synthesis with differentiable rasterization. This allows supervising our model with the combination of photometric loss and pointmap regression loss, enhancing both photorealism and structural accuracy. Given a sequence of ordered images, PreF3R incrementally reconstructs the 3D Gaussian field at 20 FPS, therefore enabling real-time novel-view rendering. Empirical experiments demonstrate that PreF3R is an effective solution for the challenging task of pose-free feed-forward novel-view synthesis, while also exhibiting robust generalization to unseen scenes.
title PreF3R: Pose-Free Feed-Forward 3D Gaussian Splatting from Variable-length Image Sequence
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2411.16877