BERT-hLSTMs: BERT and Hierarchical LSTMs for Visual Storytelling
This work aims to improve the coherence of automatically generated stories from image sequences, which is a problem for users of visual storytelling systems. It represents an incremental improvement over existing methods.
The paper addresses the lack of coherence in visual storytelling by proposing a hierarchical framework that models sentence-level and word-level semantics separately. Their model, using BERT for embeddings and a hierarchical LSTM, outperforms most related baselines on BLEU and CIDEr.
Visual storytelling is a creative and challenging task, aiming to automatically generate a story-like description for a sequence of images. The descriptions generated by previous visual storytelling approaches lack coherence because they use word-level sequence generation methods and do not adequately consider sentence-level dependencies. To tackle this problem, we propose a novel hierarchical visual storytelling framework which separately models sentence-level and word-level semantics. We use the transformer-based BERT to obtain embeddings for sentences and words. We then employ a hierarchical LSTM network: the bottom LSTM receives as input the sentence vector representation from BERT, to learn the dependencies between the sentences corresponding to images, and the top LSTM is responsible for generating the corresponding word vector representations, taking input from the bottom LSTM. Experimental results demonstrate that our model outperforms most closely related baselines under automatic evaluation metrics BLEU and CIDEr, and also show the effectiveness of our method with human evaluation.