densecap

Generated image description in the form of coherent paragraphs using Densecap(CNN for Dense Captioning) and Hierarchical RNN and Model was able to perform as good as state-of-art in terms of metrics inclined towards human-like sentences.

Processing data produced by flickr30k_entities to use as regional description for densecap model

Code repository for the final year project focusing on improving visual dialog by reducing modality biases.