A Novel Adaptive Attention Model for Image Captioning

Image captioning is an extremely challenging work, which involves both computer vision and natural language processing. However, in recent years, with the introduction and improvement of various new technologies, it has made a great advancement. In this paper, a novel attention-based image caption model is proposed. We use Resnet harnessed by outstanding performance in convolutional neural network, as the encoder. For the decoding, LSTM network equipping with fitness of dealing with problems highly related to time series, is adopted. At each time step, the attention mechanism guides the language model to synthesize context vectors adaptively. The experimental results show that the model has good performance in various quantitative metrics as well as in human evaluation.