gaussian splatting

This study systematically compares photogrammetry and Gaussian Splatting for 3D reconstruction and novel-view synthesis, proposing a synergistic reconstruction framework that integrates their respective strengths. To this end, a multi-view real-scene dataset is constructed; the Gaussian Splatting codebase is extended to support arbitrary camera poses in Blender for high-fidelity synthetic view generation, thereby augmenting training data. Quantitative evaluation employs SSIM, PSNR, LPIPS, and USAF resolution charts across multiple dimensions. The key contribution is the first empirical validation and utilization of Gaussian Splatting–generated novel views as “pseudo-ground-truth” supervision to enhance photogrammetric inputs—yielding substantial improvements in geometric accuracy and texture fidelity. Experiments demonstrate an average 23.6% gain in detail preservation and local resolution, with particularly pronounced gains in texture-deficient and occluded regions.

To address poor reconstruction quality in occluded and undersampled regions under sparse-view settings for Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), this paper proposes the first novel view synthesis framework leveraging collaborative optimization of repetitive structures. Our method employs unsupervised instance segmentation to identify repeated elements in the scene, followed by rigid and non-rigid cross-instance registration to achieve geometric alignment. We further design a cross-instance feature fusion mechanism that shares geometric priors across instances while preserving their distinct appearances. By jointly enforcing structural consistency and appearance diversity, our framework significantly improves geometric completeness and texture fidelity in occluded and sparsely observed regions. Extensive experiments on both synthetic and real-world datasets demonstrate consistent improvements in PSNR and SSIM, validating the effectiveness of explicit repetitive-structure modeling for reconstruction under low-coverage observation conditions.

Gaussian splatting (GS) surface reconstruction suffers from limited representational capacity when using a single primitive type (e.g., ellipses or ellipsoids), hindering accurate modeling of complex geometries. To address this, we propose a hybrid-primitive Gaussian point latticization framework. Our method introduces, for the first time, differentiable geometric primitives—specifically ellipses and ellipsoids—into the GS representation. We further design a composite lattice construction strategy, a hybrid-primitive co-initialization mechanism, and an adaptive vertex pruning algorithm to enhance modeling flexibility and geometric fidelity. Experiments demonstrate that our approach significantly outperforms single-primitive baselines across diverse complex shapes, achieving consistent improvements in PSNR, SSIM, and surface geometric error metrics. This work establishes a more expressive and generalizable paradigm for GS-based surface reconstruction.

This work addresses the ill-posed problem of monocular dynamic scene view synthesis by systematically evaluating and analyzing multiple Gaussian splatting–based dynamic modeling approaches. We propose the first standardized benchmark framework tailored for monocular dynamic Gaussian splatting, featuring a controllable synthetic dataset that disentangles motion, occlusion, and illumination factors, alongside joint evaluation across diverse real and synthetic datasets. Key findings include: (1) existing methods achieve high rendering efficiency but suffer from optimization fragility; (2) incorporating motion priors—particularly motion smoothness—significantly improves robustness; and (3) performance rankings are clear on synthetic data but obscured on real-world data due to uncontrolled confounding factors. Based on these insights, we distill seven reproducible, empirically grounded design principles, providing both theoretical foundations and practical guidelines for dynamic Gaussian modeling.

Addressing the challenge of real-time multi-view stereo (MVS) reconstruction and novel view synthesis (NVS) under sparse-view settings, this paper introduces the first end-to-end feedforward 2D Gaussian splatting framework. The method directly regresses generalizable 2D Gaussian parameters, jointly optimizing geometric reconstruction accuracy and rendering quality. To enhance precision, speed, and cross-dataset generalization, it incorporates multi-view feature distillation and explicit MVS supervision, integrating pre-trained visual features. Experimental results demonstrate state-of-the-art performance on DTU (Chamfer distance), significant improvements over existing methods on BlendedMVS and Tanks and Temples, and inference speed approximately 100× faster than implicit volumetric rendering. The core contribution lies in the first differentiable, generalizable, and real-time feedforward 2D Gaussian splatting reconstruction pipeline tailored for sparse-view scenarios.

Existing general-purpose novel view synthesis methods employ fixed allocation strategies for Gaussian primitives, which struggle to adapt to spatial complexity variations across scenes, resulting in redundant resources in smooth regions and insufficient representation in detailed areas. This work proposes SplatWeaver, a framework that introduces dynamic primitive allocation within a feed-forward 3D Gaussian splatting architecture for the first time. By integrating a mixture-of-Gaussians expert model with a pixel-level routing mechanism—guided by high-frequency structural priors and enhanced through routing regularization—the method adaptively allocates Gaussian primitives according to local geometric complexity. Experiments demonstrate that SplatWeaver significantly outperforms state-of-the-art approaches using fewer Gaussians, consistently achieving higher rendering fidelity and improved detail reproduction across diverse scenes.

This work addresses the limitations of existing Gaussian splatting methods, which often suffer from multi-view inconsistencies and floating-point artifacts that hinder high-quality geometric reconstruction. To overcome these issues, the paper introduces a novel approach that rigorously models Gaussian primitives as stochastic entities and leverages their volumetric properties to construct an explicit geometric representation. This formulation enables high-fidelity depth map rendering and accurate extraction of fine-scale geometry. By establishing a geometrically interpretable theoretical foundation for Gaussian splatting and integrating multi-view geometric optimization, the proposed method achieves state-of-the-art performance in shape reconstruction accuracy and consistency, significantly outperforming existing approaches on public benchmarks.

Existing single-step feedforward networks struggle to regress static Gaussian primitives suitable for all viewing angles in novel view synthesis from pose-free images, limiting reconstruction fidelity. This work proposes a viewpoint-adaptive dynamic Gaussian splatting method that transforms static representations into a view-aware dynamic splatting mechanism. Specifically, a lightweight dynamic MLP predicts residual updates to Gaussian attributes—including position, scale, rotation, opacity, and color—conditioned on the target viewpoint coordinates. By adapting Gaussian parameters dynamically to each novel view, the method significantly enhances synthesis quality while maintaining high computational efficiency, achieving state-of-the-art fidelity with inference at 17 FPS and real-time rendering at 154 FPS.

Existing feed-forward 3D Gaussian splatting methods rely on fixed Gaussian distributions, lacking adaptability across different viewpoints. This work proposes a viewpoint-conditioned feed-forward 3D Gaussian splatting model that dynamically fuses view-agnostic embeddings with target-view-specific information through a dual-branch hypernetwork, enabling query-view-customized radiance field reconstruction. By unifying cross-view sharing and view-adaptive mechanisms while preserving feed-forward efficiency, the method significantly enhances generalization capability. Experiments demonstrate that the proposed model outperforms state-of-the-art approaches on the RealEstate10K, ACID, and DTU datasets and exhibits exceptional performance in cross-dataset evaluations.

This work addresses the inherent conflict in 3D Gaussian Splatting (3DGS) between jointly modeling appearance and geometry, where direct geometric extraction often degrades rendering quality. To resolve this, the authors propose a lightweight decoupling mechanism that assigns each Gaussian an independent geometric opacity parameter and employs an opacity-guided optimization strategy. Leveraging geometric priors from vision foundation models, this approach enables disentangled modeling of appearance and geometry. The method significantly improves both novel-view synthesis quality and geometric reconstruction accuracy across multiple datasets, demonstrating particularly strong performance in complex scenes containing transparent objects.