Oct 2025· International Journal of Computer Vision· Vol 134· 5 citations· ⚡ 1 influential· 86 references
Computer Science
TL;DR
This work proposes SaLon3R, a novel framework for Structure-aware, Long-term 3DGS Reconstruction that effectively prunes the redundant 3DGS and resolves artifacts in a single feed-forward pass, and introduces a 3D Point Transformer to overcome geometric inconsistencies caused by long-term accumulative errors.
Abstract
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 (>\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$>\!$$\end{document}10FPS), robustness, and generalization ability for long-term feed-forward Gaussian reconstruction.
Existing 3D mesh reconstruction methods from Gaussian scene representations predominantly rely on iterative optimization, resulting in slow inference and limited scalability to high-resolution inputs. In this paper, we present AnyGS2Mesh, the first feed-forward framework for directly reconstructing 3D meshes from 3D Gaussian Splatting representations with support for arbitrary input image resolutions. Our approach incorporates a Gaussian-Guided Transformer architecture that exploits explicit 3D geometric priors for efficient mesh generation. We introduce three key components: (1) a Gaussian-Guided Spatial Reasoning Transformer represents Gaussian primitives as structured 3D tokens and jointly reasons over Gaussian and image features; (2) a Streaming and Patchwise Geometry Encoder processes native-resolution views sequentially and aggregates information across variable-length view sets; (3) a Scale-Aligned Hybrid Depth Refiner uses a PatchFusion-style encoder--decoder to fuse RGB-conditioned predicted depth with Gaussian-rendered metric depth, combining fine local structures with globally consistent metric scale. The refined depth maps are integrated through TSDF fusion, followed by Marching Cubes for deterministic mesh extraction. Extensive experiments show that AnyGS2Mesh achieves state-of-the-art reconstruction quality while significantly reducing inference time compared with optimization-based baselines, enabling near-real-time, high-quality mesh reconstruction. Our results demonstrate the potential of combining Gaussian representations and feed-forward Transformer architectures for scalable 3D geometry reconstruction. The code will be made publicly available upon acceptance.
Yuxuan Song, Fan Gao, Yi-Bo Zhao et al.· 0 citations
3D scene reconstruction, modeling, and rendering are highly relevant for numerous tasks, and 3D Gaussian splatting has become a standard choice in this context. Its feed-forward variants provide fast reconstruction from sparse input views but often produce per-pixel primitives, leading to highly redundant and thus inefficient representations. We present a structure-aware merging pipeline that takes per-pixel primitives from any feed-forward method and consolidates them into a compact, content-adaptive Gaussian set while largely retaining visual quality at just $\frac{1}{20}^\text{th}$ of the Gaussians of a per-pixel method. We group spatially coherent Gaussians of similar appearance into variable-size clusters via adaptive superpixel segmentation guided by a saliency map, which allocates fine segments to textured regions and coarse segments to homogeneous areas. We compress each cluster into a compact latent representation through a learned encoder, then match and consolidate representations across views based on geometric overlap and feature similarity via a learned merger. A level-of-detail decoder then produces the final Gaussians at a controllable resolution, enabling a flexible quality-efficiency trade-off at inference. As a post-processing module, the pipeline is backbone-agnostic, leveraging the strengths of existing feed-forward methods. This leads to better and more robust quality than achieved by previous approaches that target a reduction in primitive count, while providing a highly compact representation, that can be rendered efficiently.
Tim-Felix Fassch, Jochen Kall, C. Stachniss· 0 citations
In recent years, Transformer-based 3D vision foundation models have demonstrated strong generalization in multiview geometry and scene reconstruction. However, their scalability to long-sequence, urban-scale RGB streams remains limited due to the quadratic complexity of attention mechanisms, redundant frame processing, and high GPU memory pressure caused by dense spatial tokens. Although VGGT-Long partially alleviates these issues through chunked inference and loop-closure optimization, its geometric reasoning pipeline still involves substantial redundant computation, making it difficult to balance efficiency and accuracy in long-sequence scenarios. In this paper, we revisit the computational bottlenecks in long-sequence geometric inference, with a particular focus on spatial token redundancy and cross-frame global attention. We propose π³-LEGS, an efficient geometric inference system designed for scalable long-sequence 3D reconstruction. π³-LEGS integrates three key components: (1) a π³-based permutation-equivariant inference module to enhance unordered multi-view feature aggregation; (2) a geometry-aware keyframe selection mechanism that dynamically filters low-contribution frames to reduce redundant computation; and (3) a training-free block-sparse attention strategy that adaptively generates sparse attention masks based on pooled Query–Key similarity, significantly reducing global attention overhead. Extensive experiments on the KITTI Odometry dataset demonstrate that π³-LEGS achieves an average Absolute Trajectory Error (ATE) of 26.51, improving upon VGGT-Long by 6.4%, while reducing end-to-end inference time by 15.9%. Moreover, the proposed system maintains stable performance on thousand-frame sequences without runtime failures, highlighting its effectiveness in achieving a favorable accuracy efficiency trade-off for large-scale 3D reconstruction.
Faline Fu, Xiaoli Cao, Can Tang et al.· International Conference on...· 0 citations
While feed-forward 3D Gaussian Splatting (3DGS) enables efficient 3D reconstruction, achieving high-fidelity rendering remains challenging. Existing pixel-aligned approaches suffer from spatial inflexibility and massive structural redundancy, whereas query-based methods lack 3D priors and entangle geometry with appearance, yielding blurry, pose-dependent results. To overcome these deficiencies, we propose \textbf{QuerySplat}, a feed-forward 3DGS framework driven by geometric priors and explicit appearance decoupling. Specifically, we design a dual-branch query-based decoder: the geometry branch leverages a pretrained Vision Geometric Model for spatial understanding, which intrinsically endows QuerySplat with pose-free modeling capabilities, while the appearance branch recovers high-frequency details through a dedicated pathway separated from geometric attribute regression. Extensive experiments demonstrate that QuerySplat mitigates the blurry rendering issues of early query-based models and consistently outperforms pixel-aligned approaches in rendering fidelity. On the challenging DL3DV benchmark, it achieves state-of-the-art novel view synthesis performance, with average PSNR gains of 2.30 dB and 1.04 dB over the best pose-free and pose-required baselines, respectively. Project Page: https://inspatio.github.io/querysplat.
Yinglong Li, Donghui Shen, Xiaoyu Zhang et al.· 0 citations
Few-view surface reconstruction recovers the visible surfaces of a scene from a few posed RGB images, providing the 3D models that robots need to explore and interact online. On mobile platforms, the reconstruction must be fast and geometrically accurate while keeping a small memory footprint to ensure safe and efficient operation. 3D Gaussian Splatting (3DGS) offers a high-fidelity scene representation, but building it from a few views is ill-posed, as many distinct surfaces reproduce the same images, making traditional photometric methods prone to"floater"artifacts. End-to-end methods resolve the ambiguity by regressing splats with large, usually Transformer-based, networks that require heavy compute and memory while generalizing poorly to new scenes. We propose G2SR, which exploits a well-posed core of the task: given cross-view 2D splat correspondences, 3D splats follow analytically from multi-view geometry. G2SR employs a lightweight neural frontend to detect and track 2D Gaussian splats on the image plane and an analytic backend to triangulate each into a metric-scale 3D splat. On ScanNet, Replica, and DTU, G2SR matches or exceeds the geometric accuracy of state-of-the-art end-to-end methods while running at 69-89 reconstructions per second within 203 MB of GPU memory (5-107x less) for 2- and 3-view inputs at 384 x 512 resolution, offering a practical path to online Gaussian-based surface reconstruction.
Dasong Gao, Vivienne Sze, S. Karaman· arXiv.org· 0 citations
We present FlexSplat, a feed-forward framework for novel view synthesis (NVS) from uncalibrated, object-centric multi-view image collections. A recent line of query-based methods reconstructs a compact set of 3D Gaussians by treating them as transformer queries that are refined with multi-view deformable attention; these methods, however, assume that camera poses are given. FlexSplat removes this assumption: a geometry transformer is trained jointly with the Gaussian decoder to predict per-image camera parameters and depth, which in turn ground a depth-guided Gaussian parameterization and a multi-view deformable cross-attention that aggregates evidence across all input views into a single, view-consistent set of primitives. An uncertainty-weighted depth-consistency objective lets the jointly trained geometry adapt to the reconstruction task, while the cross-view consensus formed during decoding absorbs the residual error of the estimated cameras and depth. The representation uses a compact Gaussian budget that is decoupled from the input resolution - unlike pixel-aligned methods, the primitive count does not grow with the image grid - and is not dictated by the number of views. On ShapeNet-SRN and Google Scanned Objects (GSO), FlexSplat matches or approaches posed state-of-the-art reconstructors while requiring neither camera poses nor ground-truth depth, and matches the best perceptual (LPIPS) quality among the compared methods on GSO. Our results indicate that a jointly trained geometry front-end is sufficient to bring calibration-free operation to query-based Gaussian reconstruction while staying within 0.7 dB PSNR of posed methods and matching their perceptual quality.
Amir Sabbaghziarani, Hanting Ye, Maria Gorlatova et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
