🚀 A refined tactile dataset for surface defect segmentation in machine tactile perception
This repository provides a refined and finely annotated version of the HUST-Tac dataset, specifically designed for surface defect segmentation tasks in machine tactile sensing.
The dataset contains 2,364 tactile image samples, covering four defect categories commonly encountered in industrial inspection scenarios.
| Category | Description | Samples |
|---|---|---|
| 🟢 Normal | No defect present | 984 |
| 🔵 Bump | Surface protrusion | 408 |
| 🟡 Pit | Surface indentation | 486 |
| 🔴 Scratch | Linear surface defect | 486 |
| Total | — | 2364 |
- 🖼️ Image format: JPG
- 📐 Resolution: 128 × 128
- 🏷️ Filename:
<defect_type>_<index>.jpg - 🎭 Segmentation mask: PNG file with the same filename
Each mask provides pixel-wise annotations, enabling seamless integration into supervised segmentation pipelines.
This dataset is designed to support research on:
- 🤖 Machine tactile perception
- 🧠 Deep learning–based defect segmentation
- 🧪 Micro-defect detection beyond visual sensing
- 🏭 Industrial surface inspection
📢 We will continuously update this repository after the formal publication of our related paper, including improved annotations, documentation, and benchmark results.
Our goal is to foster progress in the field of tactile-based surface defect detection and segmentation.
📌 If you use this dataset in your research, we kindly ask that you cite our corresponding paper.
📄 @article{xi2026vis2tac, title={Vis2Tac: Residual feature--mediated cross-modal mapping learning framework for surface micro-defect detection}, author={Xi, Zerui and Gao, Yiping and Li, Xinyu and Gao, Liang}, journal={Pattern Recognition}, pages={113728}, year={2026}, publisher={Elsevier}}
If you find this dataset useful, please consider starring ⭐ this repository to support ongoing development!
Keywords:
Machine Tactile Perception · Surface Defect Segmentation · GelSight · Industrial Inspection