Data annotation
Annotated surveillance imagery and video frames for person, vehicle, object in crowded, and occluded conditions. Labeled manufacturing images for defect detection, including scratches, cracks, dents, missing parts, and alignment issues using bounding boxes and segmentation masks. Performed automotive image annotation for autonomous driving datasets, labeling vehicles, pedestrians, cyclists, traffic signals, and lane markings across complex road environments. A big part of the project was keeping labels consistent across similar scenes, especially when multiple individuals or objects overlapped or moved through the same frame. The work required very close visual inspection and strict adherence to defect taxonomy guidelines, since small labeling mistakes could affect the model’s ability to distinguish between acceptable variation and true production faults.