Scholarlink

<p>Recent advances in 3D Gaussian Splatting (3DGS) have enabled feed-forward, on-the-fly reconstruction of sequential input views. However, existing methods often predict per-pixel Gaussians and combine Gaussians from all views as the scene representation, leading to substantial redundancies and geometric inconsistencies in long-duration video sequences. To address this, we propose <b>SaLon3R</b>, a novel framework for <b>S</b>tructure-<b>a</b>ware, <b>Lon</b>g-term <b>3</b>DGS <b>R</b>econstruction. Our method eliminates redundancy by introducing compact anchor primitives as a replacement for per-pixel Gaussians. These primitives are derived through a differentiable, saliency-aware Gaussian quantization process designed to preserve fidelity while ensuring a compact representation. Specifically, a foundational 3D reconstruction model is employed to predict a saliency map encoding regional geometric complexity. Guided by this saliency map, we compress redundant Gaussian primitives into compact anchors by prioritizing high-complexity regions. Furthermore, we introduce a 3D Point Transformer to overcome geometric inconsistencies caused by long-term accumulative errors. It refines attributes and saliency of the anchor primitives leveraging the learned spatial structural priors in 3D space. Without known camera parameters or test-time optimization, our approach effectively prunes the redundant 3DGS and resolves artifacts in a single feed-forward pass. Experiments on multiple datasets demonstrate our approach outperform state-of-the-arts on both novel view synthesis and depth estimation, while exhibiting superior efficiency (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(&gt;\!\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo>&gt;</mo> <mspace width="-0.166667em" /> </mrow> </math></EquationSource> </InlineEquation><b>10</b> <Emphasis FontCategory="NonProportional">FPS</Emphasis>), robustness, and generalization ability for long-term feed-forward Gaussian reconstruction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SaLon3R: Structure-Aware Long-Term Feedforward 3D Reconstruction from Unposed Images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年8月5日

查看全文
  • Jiaxin Guo,
  • Tongfan Guan,
  • Wenzhen Dong,
  • Wenzhao Zheng,
  • Wenting Wang,
  • Yue Wang,
  • Yeung Yam,
  • Yun-Hui Liu

摘要

Recent advances in 3D Gaussian Splatting (3DGS) have enabled feed-forward, on-the-fly reconstruction of sequential input views. However, existing methods often predict per-pixel Gaussians and combine Gaussians from all views as the scene representation, leading to substantial redundancies and geometric inconsistencies in long-duration video sequences. To address this, we propose SaLon3R, a novel framework for Structure-aware, Long-term 3DGS Reconstruction. Our method eliminates redundancy by introducing compact anchor primitives as a replacement for per-pixel Gaussians. These primitives are derived through a differentiable, saliency-aware Gaussian quantization process designed to preserve fidelity while ensuring a compact representation. Specifically, a foundational 3D reconstruction model is employed to predict a saliency map encoding regional geometric complexity. Guided by this saliency map, we compress redundant Gaussian primitives into compact anchors by prioritizing high-complexity regions. Furthermore, we introduce a 3D Point Transformer to overcome geometric inconsistencies caused by long-term accumulative errors. It refines attributes and saliency of the anchor primitives leveraging the learned spatial structural priors in 3D space. Without known camera parameters or test-time optimization, our approach effectively prunes the redundant 3DGS and resolves artifacts in a single feed-forward pass. Experiments on multiple datasets demonstrate our approach outperform state-of-the-arts on both novel view synthesis and depth estimation, while exhibiting superior efficiency ( \(>\!\) > 10 FPS), robustness, and generalization ability for long-term feed-forward Gaussian reconstruction.