Zero-shot learning with Fast RCNN and Mask RCNN through semantic attribute Mapping

Apply

Zero-shot learning with Fast RCNN and Mask RCNN through semantic attribute Mapping

Journal Title: International Journal of Engineering and Science Invention - Year 2018, Vol 7, Issue 5

Abstract

Machine learning applications increasingly address large and diverse data sets. While this leads to an abundance of some classes of data, it also uncovers objects that defy categorization in previously seen classes. Learning from sparse data set of images is challenging. The task of classifying samples taken from categories where no training examples exist is known as zero-shot learning. This problem has interest both from the practical standpoint of automatically labelling novel items, thereby saving the time needed to retrain classifiers and from the scientific standpoint of understanding how humans perform the task. Zero shot learning is a hierarchical approach to image classification problems that have no training examples. Zero shot learning is a combination of hierarchical and attribute based classification methods; the former gives a projected position in an established hierarchy, while the latter provides a ranked listing of potential classes and their estimated probabilities. Here we are using two hierarchical methods, Fast RCNN and Mask RCNN. Finally we compare the performance of both and will find the most efficient one.

Authors and Affiliations

Jilsha P J, Chinchu Krishna

Keywords

Experimental Investigations on Performance Evaluation of Four Stroke Copper Coated SI Engine With Methanol Blended Gasoline With Catalytic Converter-A Review

The performance evaluation of four stroke spark ignition engine with gasoline and alcohol (methanol and ethanol) blended gasoline was reviewed and research gaps were identified. This paper reports performance evaluation...

EP ID EP397093
DOI -
Views 105
Downloads 0