← 4D Vision

3D/4D Tracking

Dense point tracking across frames, scene flow estimation, and unified geometry-plus-tracking models — from pixel-level correspondence to world-centric 3D trajectories.

⌘K

3D Point Tracking 15+ papers

Model / MethodFull TitleVenueYear
OmniMotionTracking Everything Everywhere All at Once ICCV2023
OmniTrackFastTrack Everything Everywhere Fast and Robustly ECCV2024
SpatialTrackerTracking Any 2D Pixels in 3D Space CVPR2024
EgoPointsAdvancing Point Tracking for Egocentric Videos WACV2025
SceneTrackerLong-term Scene Flow Estimation Network T-PAMI2025
DELTADense Efficient Long-range 3D Tracking for any video ICLR2025
SeuratFrom Moving Points to Depth CVPR2025
TAPIP3DTracking Any Point in Persistent 3D Geometry arXiv2025
TrackingWorldWorld-centric Monocular 3D Tracking NeurIPS2025
SyncTrack4DSyncTrack4D: Cross-Video Motion Alignment and Video Synchronization for Multi-Video 4D Gaussian Splatting arXiv2025
V-DPMV-DPM: 4D Video Reconstruction with Dynamic Point Maps arXiv2026
DePT3RDePT3R: Joint Dense Point Tracking and 3D Reconstruction of Dynamic Scenes arXiv2025

Unified Depth, Pose & Tracking 23+ papers

Model / MethodFull TitleVenueYear
TracksTo4DFast Encoder-Based 3D from Casual Videos NeurIPS2024
Uni4DUnifying Visual Foundation Models for 4D Modeling CVPR2025
Stereo4DLearning How Things Move in 3D from Internet Stereo Videos CVPR2025
VGGTVisual Geometry Grounded Transformer CVPR2025
Zero-MSFZero-Shot Monocular Scene Flow Estimation CVPR2025
St4RTrackSimultaneous 4D Reconstruction and Tracking in the World ICCV2025
SpatialTrackerV23D Point Tracking Made Easy ICCV2025
MVTrackerMulti-View 3D Point Tracking ICCV2025
DPMDynamic Point Maps: Versatile Representation for Dynamic 3D Reconstruction arXiv2025
POMATOMarrying Pointmap Matching with Temporal Motion arXiv2025
D²USt3REnhancing 3D Reconstruction with 4D Pointmaps arXiv2025
BA-TrackBack on Track: Bundle Adjustment for Dynamic Scene Reconstruction arXiv2025
Trace AnythingRepresenting Any Video in 4D via Trajectory Fields arXiv2025
PointSt3RPoint Tracking through 3D Grounded Correspondence WACV2025
Dens3RA Foundation Model for 3D Geometry Prediction arXiv2025
FlashVGGTEfficient Visual Geometry Transformers with Compressed Descriptor Attention arXiv2025
Fin3RFine-tuning Feed-forward 3D Reconstruction Models NeurIPS2025
AMB3RAccurate Feed-forward Metric-scale 3D Reconstruction arXiv2025
HTTMHTTM: Head-wise Temporal Token Merging for Faster VGGT arXiv2025
SelfiSelfi: Self Improving Reconstruction Engine via 3D Geometric Feature Alignment arXiv2025
UniPR-3DUniPR-3D: Towards Universal Visual Place Recognition with Visual Geometry Grounded Transformer arXiv2025