MUSE: A CAD Design Benchmark with Multi-modal Ground Truth

106 CAD design cases pair natural-language specifications with multi-view ground-truth artefacts, including 2D engineering drawings and 3D rendered images. The benchmark, created by dongxiaoyu, includes hand-crafted, rubric-style evaluation guides for assessing language and multi-modal models on engineering-grade 3D design tasks. It was last updated on 2026-05-07.

Use Cases

• Benchmarking language models on interpreting natural-language design specifications.
• Evaluating multi-modal models using paired 2D drawings and 3D rendered images.
• Assessing AI performance on engineering-grade 3D design tasks with rubric-based scoring.

Strengths

• 106 distinct CAD design cases provide a defined test suite.
• Includes multi-modal ground truth with both 2D drawings and 3D rendered images.
• Each case has a hand-crafted, rubric-style evaluation guide for structured assessment.

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

huggingface

Collection Method

Likely hand-crafted or curated for benchmark purposes.

Freshness

Last updated 2026-05-07 09:59:28; freshness should be verified.