Spatial Relational Attention Using Fully Convolutional Networks for Image Caption Generation

Attention-based encoder–decoder framework has greatly improved image caption generation tasks. The attention mechanism plays a transitional role by transforming static image features into sequentia...