An Image Dataset for Appearance Defect Detection of Cigarette Packets

Each line in an annotation file corresponds to one defective object instance in the image, with the format: `class_id normalized_center_x normalized_center_y normalized_width normalized_height` All coordinate parameters are normalized to the range [0, 1], which can be directly used for training YOLO-series object detection models.

The dataset contains 9 types of appearance defects for cigarette packets. The correspondence between class IDs, English identifiers, and defect descriptions is as follows: - 0 – baoke (Logo paper cracking) - 1 – zhizhou (Logo paper wrinkling) - 2 – qiaobian (Logo paper edge warping) - 3 – zhiwu (Logo paper staining) - 4 – jiaza (Foreign matter on packet surface) - 5 – changduanke (Logo paper length mismatch) - 6 – zhapo (Packet surface crushing) - 7 – chujiao (Transparent film corner protrusion) - 8 – zhipo (Transparent film tearing) The annotation error of all object bounding boxes is controlled within 5 pixels, and the annotation accuracy meets the training requirements of industrial vision inspection models.