VisymScenes: Worldwide Video Collection for 3D Modeling from 149 Sites

258,104 images from 149 sites collected from 42 cities and 15 countries comprise this dataset for 3D modeling. Freelancers worldwide selected up to 8 interesting sites near them containing a man-made structure and collected 32 or more videos walking around each site. The dataset was curated using the Visym Collector platform for privacy-preserving visual AI data collection and was last updated on HuggingFace by doppelgangers25 in December 2024.

Use Cases

• Training 3D scene reconstruction models based on multi-view video data.
• Developing privacy-preserving computer vision techniques based on the described collection method.
• Benchmarking geographic generalization of vision models based on data from 15 countries.
• Creating synthetic environments for simulation based on real-world man-made structures.

Strengths

• 258,104 images provide a substantial visual resource.
• Data spans 149 sites across 42 cities and 15 countries, offering geographic diversity.
• Collection was structured around man-made structures, providing a focused domain.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The specific video file formats and total dataset size are unknown.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

doppelgangers25 on HuggingFace.

Collection Method

Freelancers worldwide collected videos using the Visym Collector platform.

Freshness

Last updated 2024-12-12 22:23:10.

Geography

Worldwide, covering 42 cities across 15 countries.