Abstract
This manuscript delineates the constitution of a dataset engineered to bolster research endeavours in the realm of automated inspection of seals pertinent to Medical Device Packaging. The compendium encompasses a total of 1200 images of medical device pouches, with an equitable distribution between intact seals and those exhibiting defects. Each image boasts dimensions of 3008 by 4110 by 3 pixels. Accompanying the visual data is a suite of metadata essential for the importation of images into processing software. The defects within the dataset were meticulously crafted through a process of analysis and replication of imperfections commonly encountered on manufacturing lines, with the invaluable insights of seasoned inspectors and the support of Boston Scientific. The acquisition of images was conducted within a bespoke rig, meticulously designed to mirror the conditions prevalent in manufacturing environments. In addition to the imagery, the dataset contains binary labels and segmentation masks that delineate the defects, thereby facilitating the seamless utilization of the dataset for the training and validation of Machine Learning algorithms.