Improving Image Captioning by Mimicking Human Reformulation Feedback at Inference-time

TL;DR

This paper introduces a novel inference-time feedback method for image captioning that mimics human reformulation feedback, improving caption quality especially in low-quality and non-English scenarios without retraining the core model.

Contribution

The paper proposes a new inference-time feedback approach using human reformulations, enhancing existing captioning models without retraining and demonstrating significant improvements in non-English and style transfer tasks.

Findings

Improved caption quality with reformulation feedback, especially for low-quality captions.

State-of-the-art results in German image captioning and English style transfer.

Human validation confirms specific axes of improvement.

Abstract

Incorporating automatically predicted human feedback into the process of training generative models has attracted substantial recent interest, while feedback at inference time has received less attention. The typical feedback at training time, i.e., preferences of choice given two samples, does not naturally transfer to the inference phase. We introduce a novel type of feedback -- caption reformulations -- and train models to mimic reformulation feedback based on human annotations. Our method does not require training the image captioning model itself, thereby demanding substantially less computational effort. We experiment with two types of reformulation feedback: first, we collect a dataset of human reformulations that correct errors in the generated captions. We find that incorporating reformulation models trained on this data into the inference phase of existing image captioning models results in improved captions, especially when the original captions are of low quality. We apply our method to non-English image captioning, a domain where robust models are less prevalent, and gain substantial improvement. Second, we apply reformulations to style transfer. Quantitative evaluations reveal state-of-the-art performance on German image captioning and English style transfer, while human validation with a detailed comparative framework exposes the specific axes of improvement.