arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1907.09340cs.CLcs.CVcs.LG
收藏

VIFIDEL:评价图像描述的视觉逼真度

VIFIDEL: Evaluating the Visual Fidelity of Image Descriptions

Imperial College London(帝国理工学院)

Pranava Madhyastha, Josiah Wang, Lucia Specia

摘要:

我们解决的任务,评估图像描述生成系统。我们提出了一种新的图像感知度量:VIFIDEL。它根据图像中描述的对象的标签和描述中的单词之间的语义相似性,估计生成的字幕相对于实际图像的内容的忠实性。该度量还能够在评估期间考虑人类参考描述中提到的对象的相对重要性。即使这些人类参考描述不可用,VIFIDEL仍然可以可靠地评估系统描述。该指标在两个已知数据集上实现了与人类判断的高度相关性,并且与依赖于人类参考的指标具有竞争力

英文摘要:

We address the task of evaluating image description generation systems. We propose a novel image-aware metric for this task: VIFIDEL. It estimates the faithfulness of a generated caption with respect to the content of the actual image, based on the semantic similarity between labels of objects depicted in images and words in the description. The metric is also able to take into account the relative importance of objects mentioned in human reference descriptions during evaluation. Even if these human reference descriptions are not available, VIFIDEL can still reliably evaluate system descriptions. The metric achieves high correlation with human judgments on two well-known datasets and is competitive with metrics that depend on human references

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑