Attention Layers for Automatic Image Captioning

In this blog, the problem of improper captioning to images has been studied and adding attention layers to the model has been implemented to solve it. The LSTM, which is the specialized RNN is used for the sequence of words as captions while a CNN model using InceptionV3 setting the'imagenet' weights. RNN models are used to work better with the words and CNN models work better with the images. The two are merged followed by addition of attention layers. Thus, we have defined a deep learning model that combines the CNN features with RNN with the help of an attention mechanism, thereby developing a more accurate model.