" a one-click dataset generation method for instance segmentation of electronic components"

"Due to the lack of publicly available PCBA component segmentation datasets, this study constructs an instance segmentation dataset using a self-developed AOI device. The dataset contains 2,521 high-quality component images. Each image is annotated with a fine-grained labeling strategy. Components are divided into three categories: body, leads, and solder joints. Pixel-level masks are provided for each category. The dataset is randomly split into a training set of 2,016 images and a validation set of 505 images.Furthermore, to evaluate model performance in real production scenarios and enable fair comparison with traditional AOI methods, this study uses the same self-developed equipment and annotates the data with typical horizontal-vertical rectangular boxes from traditional AOI. Thus, each component has both a pixel-level mask (as ground truth) and a rectangular box (as simulated traditional AOI output). This design allows direct comparison on the same component between: 1) model output versus fine ground truth (absolute accuracy), and 2) model output versus traditional AOI boxes (relative improvement). The rectangular box annotations serve as a comparison baseline, not as ground truth. All datasets have been made publicly available."