In Search of a Suitable Epoch Value for the Optimum Result in Image Captioning

The proposed work is carried out on retrieving the content of an image using machine learning based approach. In order to gain a deeper insight of this topic, the current state-of-the-art image caption generation technique has been considered as the baseline. The goal of this work is to find a suitable Epoch value during the training phase which can maximize the performance of a system. The proposed model is composed of six major components: 1) Data pre-processing unit; 2) Convolutional Neural Network (CNN) as an encoder; 3) At-tention mechanism; 4) Recurrent Neural Network (RNN) as a decoder; 5) Beam search to find most optimal caption; and 6) Sentence Generation and evaluation. The model is trained to maximize the likelihood of the description sentence of a given training image. Experiments on the test dataset show the accuracy of the model and the fluency of the language it learns solely from image descriptions. The challenges of this work are mentioned in detail at the last of this paper.