Paper Title
Deep Learning Approach for Fine Grained Image Recognition on Capra Dataset

Abstract
With the recent advances in Deep Learning and computation resources, the models that are based on Convolution Neural Networks (CNN) are giving state of art results in generic object recognition and even surpassed human performance on some benchmark datasets like ImageNet. But Fine Grained Image Recognition (FGIR) is still a major challenge, because of high interclass similarity between various classes and the need for domain experts to annotate the images to create large scale datasets. In this paper we evaluate the performance of Convolution Neural Networks to recognize two different species of goats. For this task we have created a new FGIR dataset called Capra with two classes Capra-Hircus (Wild/Domesticated Goats)and Ibex. Then we evaluate the performance of GoogLeNet model using various techniques like data augmentation, transfer learning etc. We were able to get an accuracy of 97.44% on our test set by fine-tuning the model which was pre-trained on ImageNet. Index Terms - Deep Learning, Convolution Neural Networks, Fine Grained Image Recognition, Capra, Fine tuning.