A Hierarchical Approach for Generating Descriptive Image Paragraphs

TLDR

Recent advances in image captioning enable novel sentence generation, yet single‑sentence captions provide only coarse detail, and dense captioning, while finer, fails to produce coherent image stories. The paper aims to generate entire paragraphs that provide detailed, unified stories describing images, overcoming limitations of single‑sentence and dense captioning. The authors develop a hierarchical recurrent neural network that decomposes images and paragraphs into constituent parts, detects semantic image regions, and reasons about language to generate descriptive paragraphs. Linguistic analysis confirms the task’s complexity, and experiments on a new image–paragraph dataset demonstrate the effectiveness of the proposed approach.

Abstract

Recent progress on image captioning has made it possible to generate novel sentences describing images in natural language, but compressing an image into a single sentence can describe visual content in only coarse detail. While one new captioning approach, dense captioning, can potentially describe images in finer levels of detail by captioning many regions within an image, it in turn is unable to produce a coherent story for an image. In this paper we overcome these limitations by generating entire paragraphs for describing images, which can tell detailed, unified stories. We develop a model that decomposes both images and paragraphs into their constituent parts, detecting semantic regions in images and using a hierarchical recurrent neural network to reason about language. Linguistic analysis confirms the complexity of the paragraph generation task, and thorough experiments on a new dataset of image and paragraph pairs demonstrate the effectiveness of our approach.