← 4D Vision

3D Scene Reconstruction

Multi-view stereo, neural surface reconstruction, feed-forward 3D methods, and large-scale scene modeling — from object-level geometry to city-scale reconstruction.

⌘K

Small-Scale Object & Scene Reconstruction 42+ papers

Model / MethodFull TitleVenueYear
COLMAPStructure-from-Motion Revisited CVPR2016
MVSNetDepth Inference for Unstructured Multi-view Stereo ECCV2019
NeuSLearning Neural Implicit Surfaces by Volume Rendering NeurIPS2021
VolSDFVolume Rendering of Neural Implicit Surfaces NeurIPS2021
PatchMatchNetPatchMatchNet: Learned Multi-View PatchMatch Stereo CVPR2021
SparseNeuSSparseNeuS: Fast Generalizable Neural Surface Reconstruction from Sparse Views ECCV2022
NeuralangeloHigh-Fidelity Neural Surface Reconstruction CVPR2023
2DGS2D Gaussian Splatting for Geometrically Accurate Radiance Fields SIGGRAPH2024
PGSRPlanar-based Gaussian Splatting for Efficient Surface Reconstruction TVCG2024
SuGaRSurface-Aligned Gaussian Splatting for Efficient 3D Mesh CVPR2024
UFOReconGeneralizable Sparse-View Surface Reconstruction CVPR2024
SceneScriptReconstructing Scenes With An Autoregressive Structured Language Model ECCV2024
EgoLifterOpen-world 3D Segmentation for Egocentric Perception ECCV2024
GaussianUDFInferring Unsigned Distance Functions through 3DGS CVPR2025
InstantSfMFully Sparse and Parallel Structure-from-Motion arXiv2025
SaLon3RStructure-aware Long-term Generalizable 3D Reconstruction arXiv2025
ZeroGSTraining 3D Gaussian Splatting from Unposed Images arXiv2025
PLANA3RZero-shot Metric Planar 3D Reconstruction NeurIPS2025
QuickSplatFast 3D Surface Reconstruction via Learned Gaussian Initialization arXiv2025
SolidGSConsolidating Gaussian Surfel Splatting for Sparse-View arXiv2025
ReTRReTR: Modeling Rendering Via Transformer for Generalizable Neural Surface Reconstruction NeurIPS2025
HyRFHyRF: Hybrid Radiance Fields for Memory-efficient and High-quality Novel View Synthesis NeurIPS2025
GeoSVRGeoSVR: Taming Sparse Voxels for Geometrically Accurate Surface Reconstruction NeurIPS2025
RobustSplatRobustSplat: Decoupling Densification and Dynamics for Transient-Free 3DGS ICCV2025
LongSplatLongSplat: Robust Unposed 3D Gaussian Splatting for Casual Long Videos ICCV2025
Liberated-GSLiberated-GS: 3D Gaussian Splatting Independent from SfM Point Clouds ICCV2025
AtlasGSAtlasGS: Atlanta-world Guided Surface Reconstruction with Implicit Structured Gaussians NeurIPS2025
MaterialRefGSMaterialRefGS: Reflective Gaussian Splatting with Multi-view Consistent Material Inference NeurIPS2025
YoNoSplatYoNoSplat: You Only Need One Model for Feedforward 3D Gaussian Splatting ICLR2026
DRGSplatDRGSplat: Depth-Regularized 3D Gaussian Splatting ICLR2026
CLoD-GSCLoD-GS: Continuous Level-of-Detail via 3D Gaussian Splatting ICLR2026
4RC4RC: 4D Reconstruction via Conditional Querying Anytime and Anywhere arXiv2026

Large-Scale Scene Reconstruction 28+ papers

Model / MethodFull TitleVenueYear
NeRF++Analyzing and Improving Neural Radiance Fields CVPR2021
NeuralReconReal-Time Coherent 3D Reconstruction from Monocular Video CVPR2021
Block-NeRFScalable Large Scene Neural View Synthesis CVPR2022
Mega-NeRFScalable Construction of Large-Scale NeRFs CVPR2022
MonoSDFExploring Monocular Geometric Cues NeurIPS2022
Zip-NeRFAnti-Aliased Grid-Based Neural Radiance Fields ICCV2023
Switch-NeRFLearning scene decomposition with mixture of experts ICLR2023
CityGaussianReal-time High-quality Large-Scale Scene Rendering ECCV2024
Scaffold-GSStructured 3D Gaussians for View-Adaptive Rendering CVPR2024
Octree-GSTowards Consistent Real-time Rendering with LOD-Structured 3D Gaussians —2024
CityGaussianV2Efficient and Geometrically Accurate Reconstruction ICLR2025
CityGS-XA Scalable Architecture for Large-Scale Scene Reconstruction arXiv2025
FastGSTraining 3D Gaussian Splatting in 100 Seconds arXiv2025
SwiftVGGTA Scalable Visual Geometry Grounded Transformer arXiv2025
MetroGSEfficient Reconstruction of High-Fidelity Large-Scale Scenes arXiv2025
ARTDECOTowards Efficient On-the-Fly 3D Reconstruction arXiv2025
ReSplatLearning Recurrent Gaussian Splats arXiv2025
LODGELevel-of-Detail Large-Scale Gaussian Splatting arXiv2025
G3SplatG3Splat: Geometrically Consistent Generalizable Gaussian Splatting arXiv2025
WorldWarpWorldWarp: Propagating 3D Geometry with Asynchronous Video Diffusion arXiv2025
NexelsNexels: Neurally-Textured Surfels for Real-Time Novel View Synthesis arXiv2025