SpatialUAV: 4,331 Visual Question-Answering Instances for UAV Spatial Intelligence

SpatialUAV is a diagnostic benchmark containing 4,331 curated visual question-answering instances for evaluating spatial intelligence in low-altitude unmanned aerial vehicle scenarios. The dataset covers 14 task types, including semantic discrimination, spatial relations, and motion understanding. It was created by Hyu-Zhang and last updated on 2026-06-29.

Use Cases

• Benchmarking spatial reasoning models for UAVs based on visual question-answering tasks.
• Training models for aerial-aerial and aerial-ground collaboration scenarios described in the dataset.
• Evaluating AI performance on motion understanding in low-altitude UAV environments.
• Developing diagnostic tools for spatial intelligence across 14 distinct task types.

Strengths

• Contains 4,331 curated visual question-answering instances.
• Covers 14 distinct task types for comprehensive spatial intelligence evaluation.
• Specifically designed for real low-altitude UAV scenarios, including collaboration and motion.

Limitations

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

Provenance

Source

Hyu-Zhang via Hugging Face

Collection Method

Likely curated from real or simulated low-altitude UAV scenarios.

Freshness

Last updated 2026-06-29 15:16:48; freshness should be verified.