Memory Augmented Recurrent Transformer

AbstractGenerating multi-sentence descriptions for videos is one of the most challenging captioning tasks due to its high requirements for not only visual relevance but also discourse-based coherence across the sentences …...