Object Localization

Object localization is a task in computer vision that aims to precisely locate instances of a specific category of objects in an image by specifying tightly cropped bounding boxes. An object proposal specifies a candidate bounding box, and if this box has sufficient overlap with the ground-truth bounding box annotated by humans, it is considered correctly localized. Object localization focuses on locating a single instance, whereas object detection emphasizes identifying all instances of the same category in the image. Object localization holds significant application value in areas such as autonomous driving, security surveillance, and medical image analysis.