木材表面缺陷数据集Oulu-Det

OULU-DET源自奥卢大学，清洗、重标注于中南大学，由839张云杉木材图像组成，使用3200K的受控卤素灯进行拍摄。每张图像都是RGB颜色，每个通道8位，尺寸为512×488像素，光照变化误差保持在1%以内。图像已标注并分割成矩形区域，每个矩形区域对应木材表面约2.5×2.5平方厘米的面积。需要注意的是，原始数据集使用ppm格式的图像，与主流数据库标准不符，标签信息以滑动窗口的方式提供，缺陷坐标位置略显粗糙，可能会影响检测性能并限制广泛使用。为解决这些问题，在得到原作者确认后，我们将ppm格式的图像转换为.jpg格式，并采用PASCAL VOC风格重新标注图像，每张图像都附带.xml注释文件。新的标注采用更适合缺陷检测的边界框格式。

OULU-DET originates from the University of Oulu, cleaned and re-label at Central South University. It comprises 839 images of spruce wood captured using a 3200 K controlled halogen lamp. Each image is in RGB color, with 8 bits per channel, sized at 512×488 pixels, and the error due to lighting variations remains within 1%. The images in the dataset are annotated and segmented into rectangular regions, with each rectangle corresponding to an approximately 2.5 × 2.5 square centimeter area of the wood surface. Notably, the original dataset utilized images in ppm format, deviating from mainstream database standards, and the label information was presented in a slide-window manner with somewhat coarse defect coordinate positions, potentially impacting detection performance and limiting widespread usage. To address these issues, following confirmation from the original author, we converted the ppm format images to .jpg format and re-annotated the images in PASCAL VOC style, with each image accompanied by a .xml annotation file. The new labeling utilized a bounding box format better suited for defect detection.