Skip to content

Photoreal City-Scale 4D Reconstruction

Photoreal city-scale 4D reconstruction sits between SLAM, mapping, neural rendering, simulation, and world modeling. The key engineering question is not whether a method can render a convincing view. The key question is what role its output can safely play: pose source, reconstruction asset, digital twin, map QA layer, simulator scene, or planner-facing map.

This page is the cross-section entry point for Gaussian, NeRF, and feed-forward reconstruction methods that matter to autonomous driving and airside digital twins. It links the SLAM method library and first-principles knowledge base to existing perception, simulation, and world-model pages.

Read This First

Use this page to separate four claims:

ClaimWhat proves itWhat does not prove it
Render qualityheld-out RGB/depth/LiDAR render metrics, novel-view inspection, dynamic-region metricsa visually impressive training-view video
Metric geometrydepth error, LiDAR residuals, surface accuracy, Chamfer/F-score, survey or RTK comparisonPSNR alone
Pose/localization qualityATE/RPE, drift per distance/time, covariance or health behavior, loop-closure validationa Gaussian map that looks aligned locally
Planner-safe occupancyexplicit free/occupied/unknown semantics, uncertainty, validation gates, safety-case evidencea radiance field or Gaussian opacity map by itself

Method Role Taxonomy

RoleMethodsLocal starting pointHow to use
Metric SLAM with Gaussian mapsGaussian-LIC, Gaussian-LIC2, GS-LIVM, VIGS-SLAM, Splat-LOAMGaussian-LIC and Gaussian-LIC2Evaluate pose, timing, calibration, and Gaussian map quality separately.
Foundation/dense visual SLAMSLAM3R, VGGT-SLAM, VGGT-SLAM++, MASt3R-SLAM, ViSTA-SLAMSLAM3R and VGGT Foundation SLAMUse for reconstruction, visual map QA, and research baselines; do not assume safety localization.
Dynamic street 4D reconstructionStreet Gaussians, DrivingGaussian, OmniRe, S3Gaussian, PVG, OG-Gaussian, EmerNeRFDynamic 4D Neural/Gaussian ReconstructionUse for digital twins, replay, map cleaning, and simulation-support assets.
Feed-forward splatting and reconstructionVGGT, AnySplat, pixelSplat, MVSplat-style methodsFeed-Forward 3D Reconstruction and SplattingUse for priors, initialization, quick reconstruction, and QA; audit hallucination and metric ambiguity.
Supporting first principlesvolume rendering, 3DGS, neural implicit SLAM, continuous-time trajectories, calibration/timingVolume Rendering, Radiance Fields, and Gaussian SplattingUse to debug why rendering, geometry, pose, and occupancy claims diverge.

Requested Method Coverage

MethodFamilyCurrent repo handlingCoverage action
Street Gaussianstracked-object dynamic 3DGSCovered in simulation/world-model pagesSummarized in the dynamic reconstruction KB taxonomy and linked from this hub.
OmniReomni dynamic urban 3DGSMissingCovered in the dynamic reconstruction KB taxonomy as full dynamic actor reconstruction.
S3Gaussianself-supervised street 3DGSCovered in simulation pagesSummarized in the dynamic reconstruction KB taxonomy.
EmerNeRFself-supervised dynamic NeRFCovered in simulation/world-model pagesSummarized in the dynamic reconstruction KB taxonomy as a NeRF-side decomposition baseline.
OG-Gaussianoccupancy-guided street GaussianThin mentionCovered in the dynamic reconstruction KB taxonomy as occupancy-guided initialization and decomposition.
PVGperiodic vibration GaussianCovered in simulation/Gaussian overview pagesSummarized in the dynamic reconstruction KB taxonomy.
DrivingGaussiancomposite surround-view 3DGSDedicated perception method pageLinked as the canonical atomic method page.
Gaussian-LICLiDAR-inertial-camera Gaussian SLAMDedicated SLAM method pageLinked as the metric Gaussian-SLAM baseline.
Gaussian-LIC2continuous-time LIC Gaussian SLAMCovered in Gaussian-LIC pageLinked as the stronger continuous-time extension.
VGGTfeed-forward visual geometry modelCovered through VGGT-SLAM pageCovered in the feed-forward KB page and linked to foundation SLAM.
AnySplatfeed-forward 3DGS from unconstrained viewsMissingCovered in the feed-forward KB page.
pixelSplatimage-pair feed-forward 3DGSMissingCovered in the feed-forward KB page.

Reading Path By Intent

IntentRead in this order
I need a pose or SLAM methodProduction LiDAR Map Localization -> Gaussian-LIC and Gaussian-LIC2 -> SLAM3R and VGGT Foundation SLAM -> Dynamic 4D Gaussian SLAM
I need a photoreal digital twinDynamic 4D Neural/Gaussian Reconstruction -> 3DGS Digital Twin Pipeline -> Neural Scene Reconstruction
I need feed-forward initialization or priorsFeed-Forward 3D Reconstruction and Splatting -> SLAM3R and VGGT Foundation SLAM
I need math and failure modesVolume Rendering, Radiance Fields, and Gaussian Splatting -> Neural Implicit SLAM and Differentiable Mapping -> Continuous-Time Trajectory Splines and GP Priors
I need occupancy or world-model contextDynamic 4D Neural/Gaussian Reconstruction -> Occupancy Flow and 4D Scenes -> DrivingGaussian -> SplatFlow

SLAM Versus Reconstruction Boundary

SLAM estimates trajectory and map state together. Dynamic Gaussian or radiance-field reconstruction may consume externally estimated poses, object tracks, camera calibrations, LiDAR priors, or dataset annotations and then optimize a renderable scene. That distinction determines whether a method belongs in the SLAM library or in reconstruction/simulation coverage.

Use this split:

OutputTreat as
Real-time pose estimate plus map update loopSLAM or odometry method
Renderable static/dynamic scene from known posesreconstruction asset
RGB/depth/LiDAR novel viewssimulator or QA artifact
Static-only layer after dynamic removalmap-cleaning candidate requiring validation
Occupancy, freespace, or traversabilityplanner-facing map only after explicit semantic and uncertainty validation

City-Scale Constraints

City-scale or airport-scale 4D reconstruction requires more than a good paper implementation.

ConstraintWhy it matters
Tiling and level of detailOne monolithic Gaussian scene will not scale to long routes, terminals, or full airport aprons.
Pose provenanceReconstruction quality depends on whether poses came from SLAM, RTK/INS, LiDAR-inertial odometry, or dataset ground truth.
Calibration provenanceCamera-LiDAR-IMU errors can become duplicated surfaces, floaters, or dynamic-object ghosts.
Dynamic-layer lifecycleParked aircraft, parked GSE, cones, chocks, and shadows need policy before they become persistent infrastructure.
Held-out-view evaluationTraining-view render quality can hide overfitting and geometry errors.
Geometric checksRGB metrics must be paired with depth, LiDAR, mesh, or survey comparisons.
Source-log lineageEvery rendered or edited asset needs route, timestamp, calibration, model, and edit provenance.

Airside And AV Deployment Cautions

  • Use Gaussian/radiance reconstructions for simulation, visual QA, map cleaning research, and operator-facing digital twins before using them as autonomy authority.
  • Keep a conservative multi-sensor localization stack as the pose authority until Gaussian map factors have calibrated health behavior.
  • Validate reflective aircraft, wet pavement, glass, night floodlights, heat shimmer, rain, sparse geometry, parked movable objects, and repeated terminal structures separately.
  • Separate observed real objects from edited or inserted simulated objects in metadata.
  • Pair photoreal scene assets with occupancy/free-space validation when planners consume the output.

Sources

Public research notes collected from public sources.