Provided by Machine Vision and Industrial Testing Laborator (MVIT Lab).
We release two metal surface defect datasets with instance-level pixel annotations: Casting Billet and Steel Pipe, as well as a Medium and Heavy Plate Surface Defect Dataset annotated in YOLO format.
- Images: 1,060 (780 defective)
- Resolution: 96×106 to 3,228×492
- Defect Types:
- Scratch
- Weld slag
- Cutting opening
- Water slag mark
- Slag skin
- Longitudinal crack
- Images: 1,227 (554 defective)
- Resolution: 728×544 (fixed)
- Defect Types:
- Warp
- External fold
- Wrinkle
- Scratch
- Images: 680 (480 defective)
- Resolution: 256×256 (fixed)
- Defect Types:
- Inclusions (In)(120 samples)
- Blocky Scale (Bs)(120 samples)
- Striated Scale (Ss)(120 samples)
- Foreign Object Embedding (Foe)(120 samples)
-
AI Pre-segmentation
Leverage SAM's predictive interface to perform batch automatic segmentation, generating initial masks based on the provided bounding box annotations and images. -
Expert Refinement
1). Identification of Suboptimal Segmentation:
Review the initial masks to identify suboptimal segmentation results through human assessment.2). Interactive Refinement:
For suboptimal results, use SAM's interactive segmentation by iteratively adding:- Positive sample points to guide the identification of the target region.
- Negative sample points to exclude interference regions.
Continuously update the segmentation results in real-time until the desired accuracy is achieved.
3). Post-processing:
- Perform threshold-based segmentation using optimal thresholds for the specific dataset.
- Apply morphological operations, including opening and closing, to smooth edges, eliminate noise, fill holes, and perform other enhancements.
Download Link(baiduyun) | Alternative links(google drive)
For dataset inquiries or collaboration opportunities: 📧 [email protected] 📧 [email protected]
Maintained by MVIT Lab @ Collaborative Innovation Center of Steel Technology, University of Science and Technology Beijing