Improving Image Captioning with Conditional Generative Adversarial Nets

TL;DR

This paper introduces a conditional GAN framework for image captioning that uses discriminators to improve caption quality and evaluation consistency, enhancing existing models and providing objective assessment tools.

Contribution

It presents a novel GAN-based approach with discriminator networks to improve and evaluate image captioning models, extending traditional RL-based methods.

Findings

Consistent improvements over all language metrics for various models.

Discriminators can serve as objective caption evaluators.

The approach generalizes and enhances existing RL-based frameworks.

Abstract

In this paper, we propose a novel conditional-generative-adversarial-nets-based image captioning framework as an extension of traditional reinforcement-learning (RL)-based encoder-decoder architecture. To deal with the inconsistent evaluation problem among different objective language metrics, we are motivated to design some "discriminator" networks to automatically and progressively determine whether generated caption is human described or machine generated. Two kinds of discriminator architectures (CNN and RNN-based structures) are introduced since each has its own advantages. The proposed algorithm is generic so that it can enhance any existing RL-based image captioning framework and we show that the conventional RL training method is just a special case of our approach. Empirically, we show consistent improvements over all language evaluation metrics for different state-of-the-art image captioning models. In addition, the well-trained discriminators can also be viewed as objective image captioning evaluators