This dataset1 contains 224x224-pixel images of defects found in metal sheets.
The data can be used to build and train an ML model that can detect defects in materials.
This repo has the following structure:
- /data/<subdirectory name>: the various subdirectories contain images of the defects indicated by their respective subdirectory name (descriptions are provided below).
- /data/defect_ten_log.csv: CSV file that maps the images to 10 nominal classification values for use in loading the data into PerceptiLabs.
The following shows a partial example of the data stored in the CSV file:
| image_path | target |
|---|---|
| silk_spot/silkspot_640.jpeg | 6 |
| water_spot/waterspot_20.jpeg | 8 |
The following table lists the 10 nominal defect classifications values and provides a brief description of each defect:
| Defect | Nominal Classification Value | Description |
|---|---|---|
| Crease | 0 | Vertical or transverse folds across a metal strip caused during the uncoiling process. |
| Crescent Gap | 1 | Defects caused by cutting, in the shape of a half circle. |
| Inclusion | 2 | Surface defects in various shapes (e.g., fish scale shape) which may be loose and easy to fall off or pressed into the metal. |
| Oil Spot | 3 | Contamination caused by mechanical lubricant that affects the product's appearance. |
| Punching | 4 | Steel strips with additional, unwanted punching holes caused by mechanical failure. |
| Rolled Pit | 5 | Periodic bulges or pits on the metal's surface often caused by work or tension roll damage. |
| Silk Spot | 6 | Wave-like plaque on the surface often caused by the uneven pressure or temperature of a roller. |
| Waist Folding | 7 | A weld line that occurs when a steel strip is changed. |
| Water Spot | 8 | Spots which occur when drying the metal during production. |
| Welding Line | 9 | Wrinkle-like folds caused by low-carbon. |
Note: see Kaggle for additional information about these defects.
Got questions, feedback, or want to join a community of machine learning practitioners working with exciting tools and projects? Check out our Community!
1 Dataset Credits: https://www.kaggle.com/zhangyunsheng/defects-class-and-location