SpaceVista: Multi-Scale Visual Spatial Reasoning Data

The SFT training data for SpaceVista, a dataset for all-scale visual spatial reasoning from millimeters to kilometers. It was created by authors including Peiwen Sun and Shiqiang Lang and was last updated on Hugging Face in May 2026. The data preview suggests categories including Tiny Tabletop, Tabletop, Indoor, Wild Indoor, and Outdoor scenes.

Use Cases

• Train visual language models for spatial reasoning tasks based on the described multi-scale focus.
• Benchmark computer vision models on scene understanding across indoor and outdoor environments.
• Develop fine-tuned models for specific scale categories like tabletop or wild indoor scenes.
• Research the transfer of spatial reasoning capabilities from millimeter to kilometer scales.

Strengths

• Dataset is explicitly designed for all-scale spatial reasoning from millimeters to kilometers.
• The description lists specific scene categories like Tiny Tabletop, Indoor, and Outdoor, suggesting structured scope.
• Last update timestamp of 2026-05-26 indicates recent maintenance.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

SpaceVista authors via Hugging Face.

Collection Method

Created as SFT (Supervised Fine-Tuning) training data for the SpaceVista project.

Freshness

Last updated 2026-05-26 13:24:59