Object Hallucination in Image Captioning
This addresses a critical reliability issue for users of image captioning systems, such as in accessibility or content moderation, but is incremental as it focuses on analysis and evaluation rather than a novel solution.
The paper tackled the problem of object hallucination in image captioning models by proposing a new image relevance metric to evaluate hallucination rates, finding that top-performing models on standard metrics do not necessarily hallucinate less and that errors are often driven by language priors.
Despite continuously improving performance, contemporary image captioning models are prone to "hallucinating" objects that are not actually in a scene. One problem is that standard metrics only measure similarity to ground truth captions and may not fully capture image relevance. In this work, we propose a new image relevance metric to evaluate current models with veridical visual labels and assess their rate of object hallucination. We analyze how captioning model architectures and learning objectives contribute to object hallucination, explore when hallucination is likely due to image misclassification or language priors, and assess how well current sentence metrics capture object hallucination. We investigate these questions on the standard image captioning benchmark, MSCOCO, using a diverse set of models. Our analysis yields several interesting findings, including that models which score best on standard sentence metrics do not always have lower hallucination and that models which hallucinate more tend to make errors driven by language priors.