21st EANN 2020, 5 -7 June 2020, Greece

Factors affecting accuracy of Convolutional Neural Network using VGG-16

Sruthi Chiramel, Jyoti Rawat, Doina Logofatu


  In this paper, performance evaluation of image data-set using 2-layer Convolutional Neural Network Architecture and transfer learning method, is studied on Fashion MNIST dataset. Fashion MNIST is a dataset of images consisting 70000 28x28 gray-scale images, associated with label of 10 classes. The area of research of this paper in transfer learning is limited to pre-trained neural network VGG-16. The accuracy and respective losses are evaluated using the two mentioned methods.The work under this paper is inspired by widely famous Image-net Large Scale Visual Recognition Challenge, although due to constraint on time, resources, a smaller data set i.e.Fashion MNIST is taken for the study. The work is dependent on Keras Functional API and Tensor Flow. With the accuracy 88.24% and loss 29.02 as per CNN model, the model can be utilised by online clothing stores to classify their articles under right category.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.