Application of Convolutional Neural Network for Image Classification on Pascal VOC Challenge 2012 dataset

In this project we work on creating a model to classify images for the Pascal VOC Challenge 2012. We use convolutional neural networks trained on a single GPU instance provided by Amazon via their cloud service Amazon Web Services (AWS) to classify images in the Pascal VOC 2012 data set. We train multiple convolutional neural network models and finally settle on the best model which produced a validation accuracy of 85.6% and a testing accuracy of 85.24%. 1. Introduction Using convolutional neural networks (CNN) for image classification has become the de facto standard largely due to the success of Krizhevsky et al. [1], Szegedy et al. [2], Simonyan and Zisserman [3] and especially He et al. [4], which is now the state­of­the­ art architecture for image classification. By stacking convolutional layers on top of each other, amongst other architectural artifacts, highly accurate models for task of image classification, detection and segmentation have been discover. These models have been able to nearly match, if not exceed human performance on certain datasets. CNNs are not without their own shortcomings though. Due to the sheer size of networks and the millions of parameters to be optimized, the reliance of high­performance systems increases dramatically. Though modern GPU­based systems are capable of perfoming the intensive computations required by a convolutional neural network, implementing a CNN is unsuitable for those without access to such a high­ performance system. Also, more complex architectures such as those by [4] take up to 2­3 weeks to train which demonstrate the inherent difficulties in training deep convolutional neural networks. However, given the fact that CNNs provide the best results compared to other image classification models and also motivated by the credibility of the performance of CNNs by [1]–[4], we train multiple