SEE-OOD: Supervised Exploration for Enhanced Out-of-Distribution Detection

A 68.5 MB dataset of 3D point clouds for manufacturing defect inspection, released by Xiaoyang Song on figshare in May 2026. It supports research into detecting new, unseen defect types that are out-of-distribution from training data. The dataset is used to validate a novel generative adversarial approach for supervised OoD sample generation.

Use Cases

• Training out-of-distribution detection models based on limited real OoD samples
• Benchmarking OoD detection methods for surface defects using 3D point cloud data
• Exploring unseen OoD space in manufacturing defect inspection scenarios
• Augmenting real OoD samples to reduce overfitting in anomaly detection systems

Strengths

• 68.5 MB dataset size indicates a substantial collection of 3D point cloud data
• Includes real-world 3D point cloud data for manufacturing defects as mentioned in the description
• Released under a permissive CC-BY-4.0 license for broad reuse

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Data may reflect bias inherent to the specific manufacturing context and defect types captured

Provenance

Source

figshare

Collection Method

Likely collected from real-world manufacturing defect inspection scenarios.

Freshness

Last updated 2026-05-12 18:33:51; freshness should be verified