BERT-hLSTMs: BERT and Hierarchical LSTMs for Visual Storytelling

TL;DR

This paper introduces a hierarchical visual storytelling model combining BERT and LSTMs to generate coherent stories from image sequences, improving upon previous methods by modeling sentence and word dependencies.

Contribution

The novel framework integrates BERT with hierarchical LSTMs to better capture sentence-level and word-level semantics in visual storytelling.

Findings

Outperforms baseline models on BLEU and CIDEr metrics.

Demonstrates improved coherence through hierarchical modeling.

Validated by human evaluation.

Abstract

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.