📊 Dataset Statistics
The HUST-Tactile dataset is officially split into training and test sets, with detailed class distributions shown below.
| Class | Training Set | Test Set | Total |
|---|---|---|---|
| Normal | 1,272 | 318 | 1,590 |
| Bump | 584 | 136 | 720 |
| Pit | 656 | 165 | 821 |
| Scratch | 656 | 162 | 818 |
| Total | 3,168 | 781 | 3,949 |
📌 Dataset Description
This dataset was collected using a GelSight tactile sensor and contains tactile data of surface micro-defects across different surfaces. It includes four defect categories: Normal, Bump, Pit, and Scratch, with a total of 3,949 samples.
The dataset is designed to support research on tactile-based surface micro-defect detection, particularly in scenarios where visual inspection is challenging.
📖 Citation
If you use this dataset in your research, please consider citing our paper:
Z. Xi, X. Li, L. Gao, and Y. Gao, “Touch: A New Paradigm Based on Machine Tactile Sensation for Surface Microdefect Detection,” IEEE Transactions on Industrial Informatics, doi: 10.1109/TII.2025.3638047.