yukangcao/Awesome-4D-Spatial-Intelligence

A curated list of awesome papers for reconstructing 4D spatial intelligence from video. (arXiv 2507.21045)

What it solves

Researchers and practitioners face a fragmented landscape of methods for reconstructing 3D and 4D scenes from video or images. This project solves that by collecting and summarising more than 500 recent studies in one place, covering everything from basic depth estimation to complex human-object interaction and physics-aware simulation.

How it works

The repository is a structured, continuously updated list of papers. It organises work into five levels, moving from low-level 3D cues (depth, camera pose, tracking) up through 3D scene components, 4D dynamic scenes, interactions between scene components, and finally the incorporation of physical laws. Each entry includes the paper title, venue, year, and links to the paper, project page, and GitHub code where available. The categories are designed to help users find relevant work at the right abstraction level.

Who it’s for

This is aimed at researchers, graduate students, and engineers working in computer vision, robotics, graphics, or any field that requires understanding or reconstructing 3D/4D scenes. It is also useful for anyone seeking a structured overview of the current state of the art in 3D and 4D scene reconstruction.

Highlights

  • Over 500 papers, continuously updated
  • Organised into five clear conceptual levels, from low-level cues to physical integration
  • Covers depth estimation, camera pose estimation, 3D tracking, 3D reconstruction, 4D dynamic scenes, human-centric interaction, and physics-based simulation
  • Each entry includes links to papers, project pages, and code repositories
  • Open to community contributions via issues and pull requests

Related

  • Project
    datadrivenconstruction/cad2data-Revit-IFC-DWG

    A suite of offline converters that turn Revit, IFC, and DWG files into open, structured formats like GLB, IFC4, Excel, and JSONL, without needing CAD licences. Includes desktop apps, command-line tools, a data viewer, and free validation/comparison utilities.

  • Project
    thp/psmoveapi

    An open-source library for Linux, macOS, and Windows that enables PC access to Sony Move Motion Controllers for 3D tracking and sensor fusion in AR/VR applications.

  • Project
    zju3dv/manhattan_sdf

    Manhattan‑SDF is a CVPR‑2022 research code that learns a neural signed‑distance‑function for indoor 3‑D reconstruction, adding a Manhattan‑world (axis‑aligned) regulariser to improve geometry quality. It provides conda‑based installation, scripts for training on ScanNet (or custom data), mesh extraction, and evaluation against many baselines.

  • Project
    reality-opened/openreality

    OpenReality converts phone videos into persistent 3‑D scenes and exposes them to LLM assistants via 41 MCP tools for measurement, navigation, robot‑training export, and more, with a self‑hostable backend and BSD‑2‑Clause code (non‑commercial model license).