HARDLab+

01 / SPATIAL INTELLIGENCE

WolfWorld

What if a world could begin with a video?

Explore how ordinary smartphone footage can become a coherent 3D scene that people can navigate in a browser. Bring capture, reconstruction, and interactive rendering into one accessible workflow.

Research brief Open current demo

RESEARCH BRIEF

What if a world could begin with a video?

Smartphone video to 3D reconstruction. Turn everyday mobile capture into immersive, explorable worlds.

This HARD Lab direction studies the problem as an end-to-end system rather than as a single model demo. Current related work provides useful building blocks and baselines [1] [2] [3] [4] [5]; the lab’s goal is to connect them into measurable, reproducible research artifacts.

01

Multi-view video understanding and scene reconstruction

02

Efficient representations of geometry and appearance

03

Interactive, browser-based exploration of reconstructed spaces

PROPOSED WORKFLOW

Build, measure, iterate.

01

Capture

Collect smartphone video from different viewpoints to observe the same space from multiple perspectives.

02

Reconstruct

Investigate methods for recovering camera motion, scene geometry, and appearance from the captured observations.

03

Explore

Package the resulting scene for responsive, interactive viewing in the browser.

STUDENT ENTRY POINTS

Ways to start contributing.

  • Multi-view video understanding and scene reconstruction
  • Efficient representations of geometry and appearance
  • Interactive, browser-based exploration of reconstructed spaces
USEFUL SKILLSComputer vision3D graphicsWeb developmentMachine learning

RELATED WORK / 2026-09-10

Selected papers & citations.

Five primary-source papers selected to frame this direction. Links point to the authors’ arXiv records; these are external works, not HARD Lab publications.

Download BibTeX

5 selected papers · newest first by initial submission

01
2026Research preprintarXiv:2608.15260

VGGT-Align: Bridging Local Reconstruction and Global Consistency for Long-Sequence 3D Reconstruction

Wei Zhang, Yihang Wu, Songhua Li, Qi Wang

Studies scale drift and global consistency when reconstructing long sequences in chunks, using geometric anchoring and test-time adaptation.

Why it matters here. A candidate comparison for long phone captures: test alignment drift rather than judging isolated, attractive frames.

The authors list ACM Multimedia 2026. Its November event is forthcoming as of this reading-list check; the link here cites the arXiv manuscript.

Abstract PDF
Full citation & BibTeX

Wei Zhang et al. (2026). VGGT-Align: Bridging Local Reconstruction and Global Consistency for Long-Sequence 3D Reconstruction. arXiv:2608.15260. https://doi.org/10.48550/arXiv.2608.15260

@misc{hardlab_vggt_align,
  title = {{VGGT-Align: Bridging Local Reconstruction and Global Consistency for Long-Sequence 3D Reconstruction}},
  author = {Wei Zhang and Yihang Wu and Songhua Li and Qi Wang},
  year = {2026},
  eprint = {2608.15260},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2608.15260},
  url = {https://arxiv.org/abs/2608.15260},
  note = {External related work; metadata checked 2026-09-10}
}
02
2026Research preprintarXiv:2603.27222

HD-VGGT: High-Resolution Visual Geometry Transformer

Tianrun Chen, Yuanqi Hu, Yidong Han, Hanjie Xu, Deyi Ji, Qi Zhu, Chunan Yu, Xin Zhang, Cheng Chen, Chaotao Ding, Ying Zang, Xuanfu Li, Jin Ma, Lanyun Zhu

Combines lower-resolution global geometry with higher-resolution refinement instead of applying the full transformer at every fine-scale location.

Why it matters here. Useful when small geometric details matter but GPU memory limits the number and resolution of input frames.

Abstract PDF
Full citation & BibTeX

Tianrun Chen et al. (2026). HD-VGGT: High-Resolution Visual Geometry Transformer. arXiv:2603.27222. https://doi.org/10.48550/arXiv.2603.27222

@misc{hardlab_hd_vggt,
  title = {{HD-VGGT: High-Resolution Visual Geometry Transformer}},
  author = {Tianrun Chen and Yuanqi Hu and Yidong Han and Hanjie Xu and Deyi Ji and Qi Zhu and Chunan Yu and Xin Zhang and Cheng Chen and Chaotao Ding and Ying Zang and Xuanfu Li and Jin Ma and Lanyun Zhu},
  year = {2026},
  eprint = {2603.27222},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2603.27222},
  url = {https://arxiv.org/abs/2603.27222},
  note = {External related work; metadata checked 2026-09-10}
}
03
2025Conference paper · arXiv versionarXiv:2503.11651

VGGT: Visual Geometry Grounded Transformer

Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, David Novotny

Predicts camera parameters, depth, point maps, and point tracks from multiple images in a unified feed-forward model.

Why it matters here. A core geometry baseline for a phone-video-to-scene pipeline, before reconstruction cleanup and browser delivery.

CVPR 2025 is listed on the primary record.

Abstract PDF
Full citation & BibTeX

Jianyuan Wang et al. (2025). VGGT: Visual Geometry Grounded Transformer. arXiv:2503.11651. https://doi.org/10.48550/arXiv.2503.11651

@misc{hardlab_vggt,
  title = {{VGGT: Visual Geometry Grounded Transformer}},
  author = {Jianyuan Wang and Minghao Chen and Nikita Karaev and Andrea Vedaldi and Christian Rupprecht and David Novotny},
  year = {2025},
  eprint = {2503.11651},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2503.11651},
  url = {https://arxiv.org/abs/2503.11651},
  note = {External related work; metadata checked 2026-09-10}
}
04
2023Foundational work · arXiv versionarXiv:2312.14132

DUSt3R: Geometric 3D Vision Made Easy

Shuzhe Wang, Vincent Leroy, Yohann Cabon, Boris Chidlovskii, Jerome Revaud

Uses image-pair point maps and global alignment to recover scene geometry without requiring camera calibration as an input.

Why it matters here. Establishes the point-map reconstruction perspective and a useful comparison for difficult camera trajectories.

Abstract PDF
Full citation & BibTeX

Shuzhe Wang et al. (2023). DUSt3R: Geometric 3D Vision Made Easy. arXiv:2312.14132. https://doi.org/10.48550/arXiv.2312.14132

@misc{hardlab_dust3r,
  title = {{DUSt3R: Geometric 3D Vision Made Easy}},
  author = {Shuzhe Wang and Vincent Leroy and Yohann Cabon and Boris Chidlovskii and Jerome Revaud},
  year = {2023},
  eprint = {2312.14132},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2312.14132},
  url = {https://arxiv.org/abs/2312.14132},
  note = {External related work; metadata checked 2026-09-10}
}
05
2023Foundational work · arXiv versionarXiv:2308.04079

3D Gaussian Splatting for Real-Time Radiance Field Rendering

Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, George Drettakis

Represents a scene with optimized anisotropic 3D Gaussians and renders it with a visibility-aware splatting method.

Why it matters here. A rendering representation to evaluate after geometry estimation—not, by itself, a complete phone-video reconstruction system.

Abstract PDF
Full citation & BibTeX

Bernhard Kerbl et al. (2023). 3D Gaussian Splatting for Real-Time Radiance Field Rendering. arXiv:2308.04079. https://doi.org/10.48550/arXiv.2308.04079

@misc{hardlab_gaussian_splatting,
  title = {{3D Gaussian Splatting for Real-Time Radiance Field Rendering}},
  author = {Bernhard Kerbl and Georgios Kopanas and Thomas Leimkühler and George Drettakis},
  year = {2023},
  eprint = {2308.04079},
  archivePrefix = {arXiv},
  doi = {10.48550/arXiv.2308.04079},
  url = {https://arxiv.org/abs/2308.04079},
  note = {External related work; metadata checked 2026-09-10}
}