Comparison of Deep Learning Models for Automatic Image Descriptors

Image description is a task which combines the methods like Natural Language Processing, Artificial Intelligence and Computer Vision, which aims to generate contextually and semantically correct descriptions for an image dataset. The majority of the most recent methods rely on neural networks for their execution. Different Recurrent Neural Network (RNN) models are utilised to construct models for the language, and different already developed Convolutional Neural Network (CNN) models are employed for features extraction. In this paper, the main focus is on the most widely used datasets, approaches, benchmarks and the evaluation metrics for generating image descriptions. The paper involves a comparative analysis of various models for generating the descriptions and the results obtained on the Flickr8k dataset has been analyzed for them. In addition, the advantages and disadvantages of these models are being investigated.