Novel Mathematical Foundation for Automated Annotation of Massive Image Data Sets

Modern Artificial Intelligence (AI) solutions generally employ carefully-crafted Neural Networks (NNs) that require extensive human effort to perform detection, identification, and annotation on each image to create training datasets. AI tools are desired that are optimized for object identification and annotation across diverse families of image data, are reliable and robust, not dependent on extensive training demands, and are applicable to objects of interest for both government and commercial concerns. AI outputs that are explainable and more “lightweight” to human users are needed to overcome these limitations. To remove the demand for human annotation for object detection, Skyward, Ltd. (Skyward), proposes a methodology to produce an Automated Annotation Toolkit (AAT) that takes advantage of object features to reduce labeling demands and sensitivity to view. The AAT incorporates explainability and interpretability methods to provide intuitive outputs for end users.