---

Sub-domains

Sub-domain	Focus	Papers
Geometry & Depth	Depth estimation, camera pose, geometric understanding	100+
3D/4D Tracking	Dense 3D/4D point tracking, optical flow, correspondence	80+
Reconstruction	Multi-view and monocular scene reconstruction	80+
Dynamic Scenes	4D dynamic scenes via deformable NeRFs and 4DGS	80+
Human-Centric	Human pose, shape, motion capture, avatars	80+
Physics-Based	Physics simulation, dynamics prediction, material estimation	80+

---

Key Research Areas

Geometry & Depth Estimation

Monocular and multi-view depth estimation, camera pose prediction, and geometric understanding. From classical stereo to learning-based approaches using foundation models.

Key works: DPT, MiDaS, ZoeDepth, Depth Anything, Metric3D, UniDepth, DepthCrafter

3D/4D Point Tracking

Dense correspondence estimation across video frames in 3D space. Long-range tracking, scene flow, and 4D spatiotemporal correspondence.

Key works: TAP-Vid, TAPIR, CoTracker, SpatialTracker, SceneTracker, DOT

Scene Reconstruction

Multi-view and video-based 3D/4D scene reconstruction. Neural and Gaussian approaches to dense geometry recovery.

Key works: DUSt3R, MASt3R, MonST3R, Spann3R, Fast3R, SfM-free approaches

Dynamic Scenes (4D)

Modeling and rendering of dynamic 4D scenes using deformable neural radiance fields, 4D Gaussian Splatting, and temporal extensions.

Key works: D-NeRF, 4D Gaussian Splatting, Deformable 3DGS, 4DGS, SC-4DGS, Shape of Motion

Human-Centric Motion

Human body/hand/face reconstruction, motion capture, avatar generation, and temporal modeling of human performance.

Key works: SMPL-X, WHAM, 4D-Humans, TokenHMR, GVHMR, HumanGaussian, GaussianAvatar

Physics-Based Simulation

Physics-aware neural simulation, dynamics prediction, material estimation, and physics-informed learning for 4D understanding.

Key works: PhysDreamer, PhysGen, Physics3D, PIE-NeRF, Spring-Gaus, PAC-NeRF

---