Hugging Face's logo

Datasets:
hcs-tho
/
Streaming4DGS-outputs

Modalities:
3D
Image
Formats:
imagefolder
Size:
10K - 100K
Tags:
4d-gaussian-splatting
dynamic-scene-reconstruction
neural-3d-video
Libraries:
Datasets
License:
Dataset card Data Studio Files
xet
 Community
1

You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

These are research outputs derived from the Neural 3D Video (N3V) dataset (CC BY-NC 4.0). Access is approved manually. Use is restricted to non-commercial research.

Log in or Sign Up to review the conditions and access this dataset content.

Streaming4DGS outputs

Updated: 2026-10-25 · Status: ARCHIVED (the Streaming4DGS line is frozen)

Run outputs of the Streaming4DGS research line. The code and the paper source live in https://github.com/tho-kn/Streaming4DGS. That repository cites these files by their repository-relative paths, which this dataset keeps unchanged.

Contents

  • outputs/baselines/streaming_4dgs/: baseline runs on six N3V scenes: 3DGStream (per-frame NTC models and point clouds), 4C4D (final 30k-iteration and best checkpoints, point clouds, held-out renders and metrics), MoRGS (native full-scene run), a static baseline, and the msweep_* and baseline_all_v1 streaming-update states.
  • outputs/baselines/trackersplat/: TrackerSplat runs on coffee_martini (full 300-frame CoTracker3 run and dense-initializer controls).
  • temp/experiment_outputs/streaming_4dgs/ and i3da_runs/: smaller diagnostic, admission, and publication runs.

Excluded on purpose

  • Copies of the raw N3V frames and videos (staging folders, gt image folders, images* folders, cam*.mp4). Get them from the original dataset.
  • Third-party model weights (Depth Anything V2, DUSt3R) and the TrackerSplat official release. Get them from their authors.
  • COLMAP dense-stereo workspaces, normal/depth maps, and SQLite databases, which the staging scripts regenerate.
  • Failed and interrupted attempts and intermediate 7k-iteration checkpoints.

Licence and attribution

The outputs are derived from the Neural 3D Video dataset by Li et al., "Neural 3D Video Synthesis from Multi-view Video" (CVPR 2022), which is licensed under CC BY-NC 4.0. This dataset uses the same licence: non-commercial use only, with attribution to the N3V authors. The baseline methods (3DGStream, 4C4D, MoRGS, TrackerSplat) keep their own licences.
Downloads last month
55
Total file size:
332 GB