DeepCap: A Deep Learning Model to Caption Black and White Images

Captioning of colored images has been around for quite some time now, it uses object detection and the spatial relation between the objects to generate captions. There have been numerous approaches to caption colorized images in the past, but there have been a very few. In this paper we present an approach to caption Black and white images without any attempt of colorization. We have used transfer learning to implement Inception V3, a CNN model developed by Google and a runner up in the ImageNet image classification challenge, to generate captions from Black and white images achieving an accuracy of 45.77% on the validation set.