scieee AI-readable full text Open interactive document viewer

AI-assisted image and video annotation with BIIGLE

Zurowietz, Martin; Stoye, Jens; Nattkemper, Tim W.

Full text

AI-assisted image and video annotation with BIIGLE BIIGLE What is BIIGLE? Image and video data is essential for biodiversity research and the earth sciences. This data is interpreted and analysed via image and video annotation, i.e. regions of interest and objects are marked and described. Annotating large image and video datasets is very time-consuming. Thus, specialized software is necessary to make this process as e�cient as possible. BIIGLE is such a software and provides tools for manual and AIassisted annotation, collaboration, teaching and more. sea star A polygon image annotation BIIGLE is open source software and openly developed at GitHub. A public BIIGLE instance is operated at Bielefeld University, supported by the de.NBI cloud and part of the NFDI4Biodiversity service catalog. The instance is available at biigle.de and used by thousands of users worldwide. biigle.de BIIGLE Community Learn more 38.000.000 5.500.0004.700 40 AI-assistance in BIIGLE [email protected] github.com/biigle The Machine learning Assisted Image Annotation (MAIA) tool helps to build a training dataset for machine learning and then performs automatic object detection on new datasets. It provides an e�cient user interface to review the detections and select labels to describe the detected objects. The LabelBOT tool suggests labels for new annotations in real time. Suggestions are based on existing annotations in the entire BIIGLE database. This greatly reduces the amount of user interactions required for manual annotation. The Magic SAM tool generates a detailed polygon annotation for an object with a single click. Combined with LabelBOT, manual user interactions are reduced to a minimum. The Largo tool o�ers powerful sorting and filtering mechanisms. Annotations can be sorted by "similarity" or by "outliers" to quickly identify and fix annotation errors. Detection Classification Segmentation Review