Manual image annotation is a tedious, yet necessary, task to collect labels that can be used for supervised machine learning algorithms. Most existing interfaces used to assign labels to pixels in an image require the user to hand outline each distinct object or visual concept in the image. This task can be quite time consuming, especially for objects of non-rectangular shape (e.g., trees or people). This report introduces a new annotation interface called SuperLabel that makes use of superpixel segmentation to automatically define object regions so the user simply has to assign a label to these regions instead of alsomanually defining them. The report provides details on how to set up and launch SuperLabel, and describes the labelingfunctionality provided by the system to label sets of images.
展开▼