RNSplat: Radar Neural Splatting for 3D Reconstruction and Novel View Synthesis From ISAR Image Sequences

HCSplat: a hybrid-constrained 3D reconstruction method for few-shot novel view synthesis

Semantic-guided 3D Gaussian splatting for sparse-view reconstruction in industrial digital twins

Novel view synthesis, which is essential for three-dimensional (3D) reconstruction, traditionally requires dense image sets and pre-calibrated camera parameters. However, in industrial digital twin applications spatial constraints often limit data acquisition to sparse views that fail to provide the feature correspondences necessary for accurate modeling. To overcome this bottleneck, a semantic-guided 3D Gaussian splatting (3DGS) framework tailored to sparse-view industrial reconstruction was introduced. Rather than relying on precise initialization and dense inputs, the proposed method couples explicit 3D Gaussian representations with the dense and unconstrained stereo 3D reconstruction (DUSt3R) end-to-end pose estimation model. The severe occlusions, typical of factory environments, are handled by incorporating Segment Anything Model 2 (SAM2) to hierarchically decompose the scene, yielding a structured representation that improves optimization stability. A probability density field-driven algorithm is subsequently applied to extract lightweight meshes directly from the optimized Gaussian point clouds. Evaluations on the MipNeRF360 benchmark and a custom industrial dataset demonstrated substantial improvements in the structural similarity index measure. By enabling robust reconstruction from limited viewpoints, this pipeline offers a practical geometric foundation for automated inspection and remote equipment monitoring.

Bundle-Adjusted Initialization for Efficient Earth Observation Gaussian Splatting

Abstract. Satellite imagery offers a distinct advantage in Earth observation by providing expansive coverage and enabling the monitoring of inaccessible regions without physical on-site intervention, serving as a significantly more cost-effective and scalable alternative to traditional aerial or ground-based surveys. The task of 3D reconstruction from multi-view satellite images has therefore been a pivotal point of research at the intersection of photogrammetry and remote sensing. Recently, novel-view synthesis techniques such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have accelerated the accuracy and speed of topographic modeling. Among these, Earth Observation Gaussian Splatting (EOGS) has emerged as a state-of-the-art approach by adapting 3DGS to handle the unique geometric and radiometric characteristics of satellite data, including Rational Polynomial Coefficients (RPCs) and varying solar conditions. However, the standard EOGS pipeline relies on stochastic initialization, where Gaussians are distributed uniformly within a volumetric bounding box, leading to high computational overhead and dependency on aggressive pruning that can inadvertently remove critical geometric features, particularly in areas with complex urban structures. To address these limitations, we propose Bundle-Adjusted Initialization for Earth Observation Gaussian Splatting, which leverages sparse point clouds from bundle adjustment as geometric priors for Gaussian initialization. Combined with an adaptive densification strategy, our method achieves faster convergence and improved DSM accuracy on the DFC2019 dataset compared to the EOGS baseline.

VISTA-GS: MVS-Guided Virtual View Augmentation for Sparse-View 3D Gaussian Splatting

Abstract. 3D Gaussian Splatting (3DGS) has emerged as a leading technique for novel view synthesis (NVS), yet its performance degrades drastically under sparse-view conditions. While existing methods have sought to address this by incorporating accurate 3D geometry via Multi-View Stereo (MVS) or LiDAR priors, the view-dependent appearance parameters (i.e., spherical harmonics) remain exclusively optimized on the limited training views, leading to severe appearance overfitting. This is the fundamental reason why these geometry-enhanced methods still fail to generalize to out-of-distribution (OOD) viewpoints with large baselines, such as lane-changing trajectories in autonomous driving. To address this limitation, we propose VISTA-GS (Virtual Image Synthesis and Training Augmentation), a framework that synergizes MVS-based dense initialization with a physically-grounded virtual view augmentation strategy. Specifically, we position virtual cameras at strategic offsets around the original viewpoints and render virtual training images with binary validity masks via alpha-blending. By computing photometric losses exclusively within valid mask regions, VISTA-GS injects explicit angular constraints into the optimization process, effectively regularizing view-dependent appearance without relying on any external generative model. Experiments on the LLFF benchmark and a real-world LiDAR-scanned dataset demonstrate that our method achieves state-of-the-art NVS quality under sparse-view settings, with particularly significant improvements on challenging OOD viewpoints.

Confidence matters: Leveraging Multi-view Geometric Priors for GS-based Reconstruction

3D Gaussian splatting (3DGS) has emerged as a widely-used tool for novel view synthesis, offering real-time rendering in a sparse representation. However, the method's reliance on structure-from-motion initialization and photometric optimization can lead to suboptimal geometric reconstruction, particularly for objects with high specularity. In this work, we investigate the integration of geometric priors, in the form of predicted normal and depth maps, into the 3DGS framework to improve the reconstruction quality. We analyze the effect of incorporating these priors into GS-based methods and our evaluation reveals that multi-view predictions, as they are done by the recent visual geometry grounded transformer (VGGT), outperform single-view alternatives. A major factor is the existence of a confidence map for the estimations, which comes as a by-product of multi-view models and which can significantly improve the effectiveness of priors by weighting each prediction appropriately. Extensive experiments on standard benchmarks show consistent improvement in reconstruction quality and significant gains in complex scenes including specular objects.

View-surface-aware PBR Gaussian splatting for sparse-view 3D reconstruction

Efficient 3D reconstruction and high-fidelity novel-view synthesis under sparse-view, real-time constraints remain challenging. Implicit methods like Neural Radiance Fields (NeRF) depend on dense data and long optimization, while explicit 3D Gaussian Splatting (3D-GS) often loses geometric-photometric consistency and editable materials under sparse inputs. We propose a 3D-GS-based framework that attaches a simplified Disney Bidirectional Reflectance Distribution Function (BRDF) and differentiable physically based rendering (PBR) shader to each surface Gaussian for interpretable material attributes. A view-surface-aware selective densification strategy enhances sparse-region alignment, and a two stage training pipeline balances global consistency and surface fidelity, with 64 incident rays per Gaussian for accurate specular reflection. On MVImgNet and DTU dataset, our method achieves consistent improvements under 6-view settings: +0.4dB Peak Signal-to-Noise Ratio (PSNR), +0.5% Structural Similarity Index Measure (SSIM), and -5.6% Learned Perceptual Image Patch Similarity (LPIPS) on MVImgNet, and +0.6dB PSNR and +2.9% SSIM on DTU, compared with InstantSplat-XL. The proposed densification further improves PSNR by 0.7dB with only 20% higher training time. Overall, our framework delivers higher geometric accuracy, improved material realism, and real-time rendering performance suitable for applications such as Simultaneous Localization and Mapping (SLAM) and relighting.