Captioning Based Image Using Euclidean Distance and ResNet-50

Past few years researchers have focused on using the attention and transformer model to access image insights and provide deep descriptions for an input image. Although these methods have shown greater improvementsover the regular methodology of using recurrent neural networks. But still, it is under the great influence of the gradient vanishing issue. Our way of tackling this issue is to minimize the use of gradient-intensive task and replace it with distance-mapping tasks. This means that our model predicts the output on the basis of “Euclidian Distance”. To accomplish this first we use pretrained neural network to extract features and then use K Nearest neighbor to cluster image with similar features together. Here the model is used to gather low level object information to generate relevant caption so that model can work on the top efficiency and also the generated caption is natural. To fulfill this requirement, we make use for our feature extractor model and clustering model to find the closest resembling image to our query image and return its most relevant captions