MulSeT: A Multi-view Spatial Understanding Benchmark

MulSeT is a benchmark dataset designed to challenge multimodal large language models (MLLMs) on spatial reasoning tasks. It requires models to integrate information from two distinct viewpoints of a 3D scene to answer questions. The dataset was created by WanyueZhang and was last updated on the Hugging Face platform in November 2025.

Use Cases

• Benchmarking model performance on multi-view spatial reasoning based on the described task structure.
• Analyzing architectural weaknesses in MLLMs for spatial understanding based on the associated research paper.
• Training models to answer questions requiring integration of information from multiple viewpoints.

Strengths

• Designed specifically for systematic analysis of model weaknesses in spatial understanding.
• Focuses on a challenging task requiring integration of information from two distinct 3D scene viewpoints.

Limitations

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

Provenance

Source

Hugging Face dataset uploaded by WanyueZhang.

Collection Method

Likely contains synthetic or curated data for benchmarking, as indicated by platform tags.

Time Range

null

Freshness

Last updated 2025-11-04 09:39:35; freshness should be verified.

Geography

null