SpaceVista Bench: Visual Spatial Reasoning Benchmark from Millimeters to Kilometers

SpaceVista-Bench is an evaluation benchmark for the SpaceVista project, focusing on all-scale visual spatial reasoning. The benchmark was created by a team including Peiwen Sun, Shiqiang Lang, and others. It was last updated on Hugging Face on 2026-05-26.

Use Cases

• Benchmarking model performance on visual spatial reasoning tasks based on the described focus.
• Training models for multi-scale visual understanding based on the dataset's stated scope from millimeters to kilometers.
• Evaluating the generalizability of computer vision models across different spatial scales.

Strengths

• The dataset is explicitly designed as an evaluation benchmark for a published research project.
• It focuses on a specific and defined task: all-scale visual spatial reasoning.

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

Freshness

Last updated 2026-05-26 14:37:22; freshness should be verified.