Anchor Boxes are predefined rectangles used in object detection models to predict the location and size of objects in an image. At inference time, multiple anchor boxes of different scales and aspect ratios are placed across the image grid. The model learns to refine these anchors to better match ground-truth objects. This technique enables efficient multi-object detection by allowing the network to localize and classify objects in a single pass, making it fundamental in architectures like Faster R-CNN and YOLO.