FUSIONET: A Scalable Framework for Image Classification
Molokwu C. Reginald, Molokwu C. Bonaventure, Victor C. Molokwu, Okeke C. Ogochukwu · viXra · 2020
Convolutional Neural Networks have become state-of-the-art methods for image classification in recent times. CNNs have proven to be very productive in identifying objects, human faces, powering machine vision in robots as well as self-driving cars. At this point, they perform better than human subjects on a large number of image datasets. A large portion of these datasets depends on the idea of solid classes. Hence, Image classification has become an exciting and appealing domain in Artificial Intelligence (AI) research. In this paper, we have proposed a unique framework, FUSIONET, to aid in image classification. Our proposition utilizes the combination of 2 novel models in parallel (MainNET, a 3 x 3, architecture and AuxNET, a 1 x 1 architecture) Successively; these relatively feature maps, extracted from the above combination are fed as input features to a downstream classifier for classification tasks about the images in question. Herein FUSIONET, has been trained, tested, and evaluated on real-world datasets, achieving state-of-the-art on the popular CINIC-10 dataset.