OmniEDA

Multimodal Datasets & Benchmarks for Engineering Intelligence

Welcome to OmniEDA! We are a research initiative from the laboratory led by Prof. Mahanth Gowda in the School of Electrical Engineering and Computer Science at The Pennsylvania State University (Penn State). Together with our collaborators, we study multimodal understanding, geometric and spatial reasoning, and tool-augmented AI for electronic and mechanical design.

🌐 Project website · 🎓 Faculty & research group · 🤗 All datasets

Datasets

Explore the data repositories behind our engineering benchmarks.

Project	Data focus	Links
OmniSch	PCB schematics and structured circuit diagrams	Dataset · Website
OmniLayout	Schematic-coupled PCB layouts, component geometry, and placements	Dataset · Website
OmniRouting	PCB routing, connectivity, and design constraints	Dataset · Website
OmniMech	Mechanical drawings and multimodal 3D part representations	Dataset · Website
OmniCAD	3D assemblies, component libraries, poses, and mate relations	Dataset · Website

Release status: The five repositories have been initialized. Dataset files are pending upload; consult each dataset card for release updates, file availability, and documentation.

Benchmarks

Our benchmarks evaluate the engineering capabilities supported by these datasets:

OmniSch — structured visual reasoning over PCB schematic diagrams.
OmniLayout — component placement, geometric legality, routability, electrical functionality, and agentic tool use.
OmniRouting — routing path planning, connectivity, design-rule compliance, and tool-assisted refinement.
OmniMech — drawing-to-3D reconstruction, parametric CAD synthesis, and annotation-grounded reasoning.
OmniCAD — component identification, 3D pose estimation, mate-graph reasoning, and agentic assembly.

A dataset provides the examples and annotations. A benchmark defines how those examples are used to evaluate a system, including tasks, evaluation protocols, metrics, and reference results. See each project website and dataset card for the applicable protocol and citation.

Access & License

Dataset access is gated with manual review. Sign in to Hugging Face, open the relevant dataset repository, and submit an access request. Requests are handled separately for each dataset; approval is required before downloading its released data.

Our datasets use the OmniEDA Dataset License, which limits use to non-commercial research and restricts redistribution. Separately identified third-party materials retain their original terms. Read the license in the repository you use.

Contact & Contributions

Please use the Community tab of the relevant dataset for questions, issues, and proposed contributions. When using a dataset or reporting benchmark results, cite its associated project publication.