Maven Torch Lighter

Maven Torch, a manufacturer of high-precision torch lighters, needed a reliable method to identify minor production defects before packaging. Their existing visual inspections were manual and inconsistent, leading to quality-control bottlenecks and missed defects.

Our Role
xLabeling provided a human-in-the-loop labeling workflow to build Maven’s first computer-vision training dataset for defect detection. Our expert labelers annotated product images to identify visual flaws such as scratches, misaligned nozzles, and color inconsistencies.

Process

• Collected and reviewed 3,000 high-resolution product images from the client’s assembly line.

• Labeled each image with bounding boxes and defect categories (“scratch,” “assembly gap,” “color deviation”).

• Applied double-review QA for consistency across the dataset.

• Delivered a structured dataset formatted for model training and validation.

Result
The labeled dataset enabled Maven Torch to train a defect-detection model that achieved over 93% accuracy in identifying surface flaws, reducing manual inspection time by nearly 40%.