Crowd-3DGS is a single-scene video-derived dataset capturing a real-world crowd protest, intended for 3D scene reconstruction and novel view synthesis. The dataset, created by author siyah1, includes high- and low-resolution image sets, COLMAP sparse reconstruction, stereo depth maps, and ready-to-run COLMAP scripts. It was last updated on March 23, 2026.
Use Cases
- Training and evaluating 3D Gaussian Splatting models based on the provided sequential-frame image sets.
- Benchmarking novel view synthesis algorithms using the real-world crowd protest scene.
- Testing stereo depth estimation methods against the included depth maps.
- Reproducing 3D reconstruction workflows with the ready-to-run COLMAP scripts.
Strengths
- Dataset is fully prepared for 3D Gaussian Splatting reconstruction, including multiple processed components.
- Includes both high- and low-resolution image sets, which likely supports different computational requirements.
- Provides stereo depth maps and COLMAP sparse reconstruction, offering multiple data representations.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Row count, column definitions, and file formats are unknown, which may limit suitability assessment.
- Data may reflect temporal or scene bias inherent to a single, specific protest event.
Provenance
- Source
- huggingface
- Collection Method
- Derived from a video of a real-world crowd protest scene.
- Freshness
- Last updated 2026-03-23 07:27:39; freshness should be verified.