Image Captioning Using Deep Transfer Learning

It has been interesting to generate captions from an image due to its real-life significance and applicability in day-to-day applications. Moreover, this phenomena is related to field of computer vision for object identification and natural language processing (NLP) for representation as well. In this research, we proposed a model which employs transfer learning approach to generate a caption from an input image. To train our model, we used Flickr8k dataset for our experiment. This dataset has images of six unique Flickr groups and in total it consists of around 8000 image and each image contains five types of captions. At first, we extracted features from the images using CNN. Subsequently these features are trained through transfer learning. Subsequently a RNN is used as a decoder for generating the language text. The trained model performs better for other datasets also.