Detects objects in images using only image-level labels, without the need for detailed annotations. Reduces manual labelling effort while enabling accurate object detection in various applications.