Ruggero1912/Trace_Captioning_Flickr30K

Flickr30k Trace Captioning Dataset是一个包含图像及其对应的空间-时间轨迹和文本描述的数据集。数据集用于评估基于区域的图像描述模型，包括基于轨迹的描述、基于区域的描述和视觉定位等任务。每个样本包含多个描述和对应的鼠标扫描模式轨迹，这些轨迹代表了在图像的任意区域上的鼠标扫描模式。数据集由Localized Narratives的注释和Flickr30k数据集的图像组成，旨在帮助研究人员理解和提高基于区域的图像描述。

The Flickr30k Trace Captioning Dataset is a resource that includes images and their corresponding spatial-temporal traces along with textual descriptions. It is designed to evaluate region-based image captioning models, including tasks such as trace captioning, region-based captioning, and visual grounding. Each sample in the dataset contains multiple descriptions paired with mouse scanning patterns that represent the movement of the mouse over arbitrary image regions. The dataset is composed of annotations from Localized Narratives and images from the Flickr30k dataset, aiming to assist researchers in understanding and improving region-based image descriptions.