CR7-DET

The CR7-DET dataset consists of surface defect images collected from a steel company in China, representing real-world industrial scenarios. It includes seven types of surface defects: inclusion, dents, oil spots, pits, punching, linear defects, and macular spots. All defects are located on the surface of cold-rolled steel plates. The dataset contains 4,140 images and 11,020 annotations. The defect categories are as follows:

0: Inclusion – Nonmetallic inclusions with spot-like or block-like surfaces, exhibiting discontinuous or continuous distribution, and appearing reddish brown, dark gray, or white in color.

1: Dents – Straight and fine grooves with varying depths on the surface of the steel plate.

2: Oil Spots – Periodic punctate spots appearing yellow or black on the surface.

3: Pits – Shallow surface depressions resembling gravures.

4: Punching – Irregular and spaced holes penetrating the surface.

5: Linear – Scratches caused by rollers or other structures.

6: Macular – Yellow and white discoloration on the surface, giving a blooming appearance.

Dataset Structure:

CR7-DET/: The main folder containing the image and label subfolders.

readme.txt: A file containing the description and instructions for using the dataset.

The dataset is organized into two main subfolders:

image/: Contains the actual steel plate images in .jpg format.

label/: Contains the corresponding annotation files in YOLO format. Each annotation file (e.g., dents1.txt) provides detailed information about the defects in the images, including defect type and location. The annotation format follows the YOLO standard, with each line in the label file corresponding to one defect in the image, with the format as follows:

Where:

class_id corresponds to the defect category (e.g., 0 for Inclusion, 1 for Dents, etc.)

x_center, y_center represent the coordinates of the center of the defect, normalized by the image width and height.

width, height are the width and height of the defect, normalized by the image width and height.

Each annotation file is linked to its corresponding image file, ensuring a one-to-one correspondence between the image and its annotations.

This dataset can be used for defect detection tasks, and users can access both the images and annotations for training and testing machine learning models.