Lithium-Ion-Cell 18650 Labeled Polarity-Aware Anomaly Dataset

This repository provides a polarity-aware industrial image dataset of cylindrical battery cells captured under realistic pick-and-place assembly conditions. The images were acquired during a representative battery pack assembly process and reflect visual variability typical of small and medium-sized enterprise (SME) manufacturing environments, including variations in placement, background context, and lighting. The dataset consists of cropped single-cell images (256x256 pixel) extracted from images of populated cell holders using a model-assisted auto-labeling and verification pipeline. Each cell image is labeled with its polarity (positive or negative) and grouped by condition (normal or anomalous), where anomalies correspond to visually observable deviations from nominal cell appearance such as damage or missing insulation. In addition to the primary dataset containing one image per unique cell, an extended variant with additional images captured under different lighting conditions and spatial positions is included and kept separate within the repository. The dataset is intended as a practical resource for research on supervised inspection and anomaly detection in assembly-related industrial vision tasks, with an emphasis on feasibility, transparency, and reuse rather than standardized benchmarking.