CrossFit-GS: Out-of-Fold Pose Estimation for COLMAP-Free 3D Gaussian Splatting

Key points

• In this work, we present CrossFit-GS, a deferred-commitment progressive framework that mitigates cumulative trajectory drift and breaks the self-reinforcing loop of distorted-map error propagation.1
• Specifically, we partition the input sequence into interleaved subsets and maintain two complementary submaps.1
• Each candidate pose is cross-refined against a frozen opposite submap that has never been optimized using the target frame, thereby providing an independent geometric reference for pose estimation.1
• Extensive evaluations across multiple benchmarks demonstrate that CrossFit-GS substantially reduces cumulative trajectory drift and consistently improves both camera pose estimation and novel-view synthesis over existing methods.1

Sources (1)

• [1]CrossFit-GS: Out-of-Fold Pose Estimation for COLMAP-Free 3D Gaussian Splatting
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 01:29 PM

  Pose-free 3D Gaussian Splatting (3DGS) aims to bypass cumbersome Structure-from-Motion (SfM) by directly reconstructing scenes from unposed image sequences.

  In this work, we present CrossFit-GS, a deferred-commitment progressive framework that mitigates cumulative trajectory drift and breaks the self-reinforcing loop of distorted-map error propagation.

Extractive summary: sentences quoted from the sources.