Plastic Shampoo Bottle Dataset for Surface Defects Detection

The dataset consists of plastic shampoo bottle images annotated with five distinct categories of surface defects: Label Adhesion on Bottle (LAOB), Dent on Bottle (DOB), Scuff on Bottle (SOB), Product on Bottle (POB), and Label Tear on Bottle (LTOB). Defects are treated as individual objects within the images. A portion of the dataset includes bottles exhibiting multiple defects simultaneously, intentionally incorporated to recognize compound, real-world defect scenarios. Each defect type is additionally assigned a quality grade: marginal or unacceptable based on the measured size and severity of the defect. The images of defective bottles were manually annotated to precisely mark the spatial locations of each defect. Every defect instance was assigned a unique class label that captured both its defect type and its severity level. Specifically, adhesion_marg and adhesion_unacc were used for the marginal and unacceptable cases of LAOB; dents_marg and dents_unacc for DOB; scuff_marg and scuff_unacc for SOB; pob_marg and pob_unacc for POB; and tear_marg and tear_unacc for LTOB. Annotation was performed using rectangular or square bounding boxes that fully enclosed the visible defect regions.

本数据集包含标注有5类不同表面缺陷的塑料洗发水瓶图像，涵盖的缺陷类别分别为：瓶身标签粘附（Label Adhesion on Bottle, LAOB）、瓶身凹痕（Dent on Bottle, DOB）、瓶身划痕（Scuff on Bottle, SOB）、瓶身残留异物（Product on Bottle, POB）以及瓶身标签撕裂（Label Tear on Bottle, LTOB）。缺陷被视作图像中的独立目标对象。
数据集包含部分同时存在多种缺陷的瓶身样本，此类样本为有意设置，用于适配真实场景中的复合缺陷识别任务。
每类缺陷还会依据其尺寸与严重程度被赋予质量等级：轻微不合格（marginal）或严重不合格（unacceptable）。存在缺陷的瓶身图像均经人工标注，以精准标记每一处缺陷的空间位置。每个缺陷实例均被分配唯一的类别标签，该标签同时涵盖缺陷类型与严重程度信息。
具体而言，LAOB的轻微与严重不合格情况分别采用adhesion_marg与adhesion_unacc作为标注标签；DOB对应dents_marg与dents_unacc；SOB对应scuff_marg与scuff_unacc；POB对应pob_marg与pob_unacc；LTOB对应tear_marg与tear_unacc。
标注工作采用完全包围可见缺陷区域的矩形或方形边界框完成。