Smart Security Image Annotation Project
Image
ClassificationParticipated in a smart security dataset annotation project for 3 months, focusing on annotating pedestrians, vehicles, and traffic signs in surveillance images. Specific tasks: Used bounding boxes to annotate pedestrians and vehicles, completing ~800 valid boxes per day on average. Applied precise polygon outlines for heavily occluded targets to ensure edge conformity. Performed three‑class classification (“visible / low‑light / no‑light”) on 2,000 night‑time images to support training. Project scale: The team annotated 12,000 images in total; I independently completed ~2,500 images (over 20% of the workload). Produced over 24,000 bounding boxes and 600+ polygons. Quality standards: Strictly followed client annotation guidelines, maintaining bounding box IoU ≥ 0.95. Participated in weekly internal quality checks, with a stable annotation accuracy of over 98.2%. Completed 3 rounds of error review and correction, becoming familiar with common error types and fix methods.