… CIDEr-D, a traditional evaluation metric for image description, … We also show that CIDEr-D
has performance hampered by … lem, we introduce CIDEr-R, which improves CIDEr-D, making …

… , the proposed improved image description generation model with attention … image
description task, and this model is better than the traditional model on all the four evaluation …

… The consensus loss can be implemented in most of the image captioning models trained by
cross entropy loss. In this section, we briefly introduce two consensus-based models built on …

… Cider: Consensus-based image description evaluation. In Proceedings of the IEEE … From
image descriptions to visual denotations: New similarity metrics for semantic inference over …

… We design a consensus based method that aligns the visual graph to language graph
through a novel generative adversarial structure network, which aims to maintain the semantic …

… In this section, we briefly review existing works, mainly 141 focusing on three categories:
RNN-based image captioning, 142 Transformer-based image captioning, and knowledge-…

… In this section we review image captioning methods and their evaluation metrics. Compared
… more evaluation metric scores, such as L2E [32] and word mover distance (WMD) [38], [39]. …

… Parikh, “Cider: Consensus-based image description evaluation,” in Proceedings of the IEEE
… Gould, “Spice: Semantic propositional image caption evaluation,” in European conference …

… the distinctiveness of image captions through training with sets of similar images. First… CIDEr
(CIDErBtw) to evaluate the distinctiveness of a caption with respect to those of similar images…

… Encouraged by the accomplishment of image captioning and machine translation tasks, the
… CIDEr is an evaluation protocol proposed to evaluate image description that uses human …