A STUDY OF ACTIVE OBJECT DETECTION OF COMPOSITE LABELLING
Abstract
This project proposes an active learning framework on object detectors. We support any imagery-based dataset and any object detector from TensorFlow’s Model zoom achieve. We put special focus on steel defect detection and composite labelling. Due to steel production having a continuous flow on steel needing to automatically be checked, we apply online learning elements to the investigation. We explore various object detectors in this unique setting of active-online learning
Download Connor Clarkson's Thesis