IRIS: Industrial Real-Sim Imagery Set for Object Detection

IRIS is a real-world dataset of 508 annotated images featuring 32 mechanical components, captured across four distinct industrial scenes. It is designed for studying sim-to-real transfer in robotic environments and includes assets for synthetic data generation. The dataset was created by moiaraya and was last updated on Hugging Face in May 2026.

Use Cases

• Training object detection models for industrial robotic grasping based on annotated real-world images of mechanical components.
• Benchmarking sim-to-real transfer performance using the provided real imagery and synthetic data generation assets.
• Developing domain adaptation techniques for robotic vision across the four distinct industrial scenes described.

Strengths

• Contains 508 annotated real-world images, providing a concrete scale for model training.
• Features 32 distinct mechanical component classes, offering variety for object recognition tasks.
• Captured across four distinct industrial scenes, which likely introduces environmental diversity.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Description metadata is limited; actual data quality and annotation details require manual inspection.

Provenance

Source

huggingface

Collection Method

Annotated images captured in industrial robotic environments; includes assets for synthetic data generation.

Freshness

Last updated 2026-05-19 13:03:52; freshness should be verified.