Zero-Shot Learning to Detect Object Instances from Unknown Image Sources

Inspired by the human capability, zero-shot learning research has been approaches to detect object instances from unknown sources. Human brains are capable of making decisions for any unknown object from a given attributes. They can make relation between the unknown and unseen object just by having the description of them. If human brain is given enough attributes, they can assess about the object. Zero-shot learning aims to reach this capability of human brain. First, we consider a machine to detect unknown object with training examples. Zero-shot learning approaches to do this type of object detection where there are no training examples. Through the process, a machine can detect object instances from images without any training examples. In this paper, we develop a dynamic system which will be able to detect object instances from an image that it never seen before. Which means during the testing process the test image will completely unknown from trained images. The system will be able to detect completely unseen objects from some bounded region of given images using zero shot learning approach. We approach to detect object instances from unknown class, because there are lots of growing category in the world and the new categories are always emerging. It is not possible to limit objects in this fast-forwarding world. Again, collecting, annotating and training each category is impossible. So, zero-shot learning will reduce the complexity to detect unknown objects.