Reference Caption

Reference captions in image captioning are crucial for evaluating the quality of automatically generated descriptions, but their limitations—such as expense and inability to capture nuanced aspects of image content—have spurred research into alternative evaluation metrics. Current efforts focus on developing both reference-based and reference-free metrics, often employing transformer-based architectures and incorporating visual features to better align automated scores with human judgment. These advancements aim to improve the accuracy and efficiency of image captioning evaluation, ultimately leading to more robust and human-like caption generation systems.