PCB Point Cloud Dataset for Semantic Segmentation Research

Point cloud data representing electronic components and their spatial relationships on printed circuit boards. The dataset was created by author Zedong Huang and published on April 4, 2026. It is a small dataset with a file size of 146.0 KB.

Use Cases

• Train and evaluate semantic segmentation models like the proposed AEC-Mamba network on PCB point clouds.
• Benchmark model robustness against challenges such as irregular point structure, small objects, and uneven density distribution described in the research.
• Apply focal loss techniques to address class imbalance present in the dataset's semantic labels.
• Study the effectiveness of Adaptive Edge Convolution for enhancing representation in boundary regions and sparse structures of point clouds.
• Test sequence modeling approaches, like Mamba modules using Hilbert space-filling curves, for capturing long-range context in unordered point sets.

Strengths

• Dataset is openly licensed under CC-BY-4.0.
• Associated research paper validates the data's use for benchmarking against methods like DGCNN on S3DIS and a PCB dataset.
• Designed to address specific industrial challenges like noise and sparse inputs.

Limitations

• The dataset is very small, with a total file size of only 146.0 KB, limiting the scope for large-scale model training.
• Specific details on the number of point clouds, points per cloud, and class distribution are unknown.

Provenance

Source

zedong huang via figshare.

Collection Method

Method of data collection or generation is not specified in the provided description.

Freshness

Last updated on April 4, 2026.