SPAR-Bench: 7,207 Spatial Reasoning QA Pairs Across 20 Tasks

SPAR-Bench contains 7,207 manually verified spatial reasoning question-answer pairs across 20 distinct tasks, released by jasonzhango in 2025. The benchmark evaluates vision-language models using single-view, multi-view, and video modalities to test spatial perception and reasoning capabilities.

Use Cases

• Evaluating VLM spatial perception across 20 task categories
• Benchmarking multi-view reasoning using multi-image QA pairs
• Testing temporal spatial reasoning within the video-based tasks

Strengths

• 7,207 manually verified records
• 20 diverse spatial task categories
• Includes three distinct visual modalities: single-view, multi-view, and video

Limitations

• Small scale of 7,207 records compared to general VLM training sets
• Specific column names and schema are not explicitly listed in the source metadata

Provenance

Source

jasonzhango (Arxiv: 2503.22976)

Collection Method

Manually verified

Freshness

Last updated April 2025.