Changing Input Shape Dimension Using VGG16 Network Model

Abstract

In computer vision, transfer learning is a common method because it helps us to quickly create accurate models. In this work, consider the outcome of the convolutional network depth with VGG16 on its accuracy in the large-scale image recognition setting. Rather than using a Convolutional Neural Network, Transfer Learning can be used on images with different image dimension inputs (CNN) and was originally trained on by using Keras to fine-tune the input from tensor dimensions. In this paper, we demonstrate how the VGG16 network handles new image input dimensions of 128x128x3 pixels from eligible VGG16 224x224x3 pixels images that are cut before the recognition is implemented. Our results show that Convolutional Neural Network can manage small datasets and can produce ideal validation accuracy of 93% from small images and better results from higher resolution images.