Abstract: The ceramic package substrate plays a crucial role in the field of integrated circuit manufacturing. Nevertheless, the lack of a well-defined dataset and benchmark, as well as the scarcity of defective ceramic package substrate samples, hinders further research and optimization of the project. To address these issues, we introduce CPS3D-Det, a high-precision 3D industrial defect detection dataset based on IC ceramic packaging substrates. All of the samples are collected from the multi-batch production stages in actual industrial production scenarios. With 1640 high spatial resolution point cloud samples (0.0025 mm) and hundreds of millions of total points. In addition, we propose a benchmark AIFAD, an end-to-end point cloud defect detection method based on auxiliary information flows. AIFAD transforms sparse convolutions into dense features to enhance the sparse backbone and does not rely on handcrafted proxies. Through comprehensive evaluations, we demonstrate the relevance and effectiveness of our dataset and benchmark. They will be released at https://github.com/dcsrgh/CPS3D-Det.

🎥 Demo

Collection system

Click here to view the demo video.

Complete multi-view videos

Click here to download the complete multi-view videos.

Data Preparation

1. Download the CPS3D-Det dataset from here. The data to be downloaded includes:

   • ImageSets (Complete).
   • labels (Complete).
   • points (training set, geometric sampling).
   • velodyne (training set).

2. Place according to the following structure.

CPS3D-Det
├── data
│   ├── cps3d-det
│   │   │── ImageSets
│   │   │── labels
│   │   │── points
│   │   │── velodyne
├── pcdet
├── tools

Evaluation

We provide the trained weight file, so you can just run with that. You can also use the model you trained.

cd tools 
bash scripts/dist_test.sh NUM_GPUS --cfg_file PATH_TO_CONFIG_FILE --ckpt PATH_TO_MODEL
#For example,
bash scripts/dist_test.sh 8 --cfg_file PATH_TO_CONFIG_FILE --ckpt PATH_TO_MODEL

Training

bash scripts/dist_train.sh NUM_GPUS --cfg_file PATH_TO_CONFIG_FILE
#For example,
bash scripts/dist_train.sh 8 --cfg_file PATH_TO_CONFIG_FILE

CPS3D-Det Dataset Description

Illustration of the acquisition process of CPS3D-Det.

Comparison results of the datasets from realistic industrial scenario.

Examples of the proposed CPS3D-Det.

Comparison between the proposed CPS3D-Det and existing mainstream 3D industrial datasets.

Methodology

Illustration of the proposed AIFAD.

Performance of AIFAD and the state-of-the-art methods on the CPS3D-Det validation set.

License

The dataset is released under the CC BY 4.0 license. All data collection processes are authorized by the relevant companies.

Acknowledgements

We thank these great works and open-source repositories: OpenPCDet, VoxelNeXt and CenterPoint.

Citation

If you find this project useful in your research, please consider cite:

@ARTICLE{11361169,
  author={Yu, Ruiyun and Zhao, Ziming and Zhen, Shi},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Auxiliary Information Flow for 3D Industrial Defect Detection on IC Ceramic Package Substrate Surfaces: Dataset and Benchmark}, 
  year={2026},
  volume={36},
  number={6},
  pages={7444-7456},
}