Spatial457: 6D Spatial Reasoning Benchmark with 10K-100K Records

Spatial457 contains between 10,000 and 100,000 image-text pairs designed for 6D spatial reasoning diagnostics. Created by researchers at Johns Hopkins University and DEVCOM Army Research Laboratory in 2025, it benchmarks the ability of multimodal models to interpret 3D orientations. The data is released under an Apache 2.0 license.

Use Cases

• Evaluating LMM performance using 6D spatial reasoning queries to assess orientation accuracy
• Benchmarking image-to-text models on their ability to resolve 3D positional relationships
• Fine-tuning vision-language models using the 6D pose features to improve spatial awareness

Strengths

• Scale of 10,000 to 100,000 records
• Apache 2.0 open-source license
• Specialized 6D spatial reasoning focus

Limitations

• Highly specialized scope may not generalize to non-spatial multimodal tasks
• Lack of documented column names in the current metadata

Provenance

Source

Johns Hopkins University and DEVCOM Army Research Laboratory

Freshness

Last updated March 2026.