Computer Vision is being leveraged everywhere today; be it Healthcare, Infrastructure, Transportation, Security, or Manufacturing. Use of Computer Vision is going to increase exponentially. This trajectory will include examining millions, or even hundreds of millions, of images, depending on how robust or how complex one wants their model to be.
While collecting large amounts of unlabelled data is already a major challenge, only a small subset of it can be labelled by humans due to the effort needed for high-quality annotation. Human effort also adds errors and inconsistency in labelling. Labelling is one of the greatest stumbling blocks to the wide adoption of Computer Vision in Enterprises.
Active Learning seems to be one of the answers to this conundrum. How can we obtain high-quality annotation with minimal supervision? What role does active learning play in this whole mechanism? How can we make it more scalable? What are the challenges in this approach?
Let us explore the role of active learning in annotation…