Image captioning via hierarchical attention mechanism and policy gradient optimization



Yan, Shiyang, Xie, Yuan, Wu, Fangyu, Smith, Jeremy S ORCID: 0000-0002-0212-2365, Lu, Wenjin and Zhang, Bailing
(2020) Image captioning via hierarchical attention mechanism and policy gradient optimization. SIGNAL PROCESSING, 167. p. 107329.

[img] Text
last_submission.pdf - Author Accepted Manuscript

Download (11MB) | Preview

Abstract

Automatically generating the descriptions of an image, i.e., image captioning, is an important and fundamental topic in artificial intelligence, which bridges the gap between computer vision and natural language processing. Based on the successful deep learning models, especially the CNN model and Long Short Term Memories (LSTMs) with attention mechanism, we propose a hierarchical attention model by utilizing both of the global CNN features and the local object features for more effective feature representation and reasoning in image captioning. The generative adversarial network (GAN), together with a reinforcement learning (RL) algorithm, is applied to solve the exposure bias problem in RNN-based supervised training for language problems. In addition, through the automatic measurement of the consistency between the generated caption and the image content by the discriminator in the GAN framework and RL optimization, we make the finally generated sentences more accurate and natural. Comprehensive experiments show the improved performance of the hierarchical attention mechanism and the effectiveness of our RL-based optimization method. Our model achieves state-of-the-art results on several important metrics in the MSCOCO dataset, using only greedy inference.

Item Type: Article
Uncontrolled Keywords: Image captioning, Hierarchical attention mechanism, Generative adversarial network, Reinforcement learning, Policy gradient
Depositing User: Symplectic Admin
Date Deposited: 10 Oct 2019 15:48
Last Modified: 19 Jan 2023 00:23
DOI: 10.1016/j.sigpro.2019.107329
Related URLs:
URI: https://livrepository.liverpool.ac.uk/id/eprint/3057723