CompGuessWhat?!

该数据集名为CompGuessWhat?!，它采用了一种多任务评估框架，旨在评估在学习神经表征过程中，关于属性扎根的质量。该数据集对与GuessWhat?!图像相关的VisualGenome场景图进行了丰富，包含了预测抽象、具体情境和位置属性的多个任务。它扩展了原始的GuessWhat?!数据集，增加了额外的语义层和属性。其任务包括基于属性的语言学习，涉及目标导向的评估和对象属性预测。

This dataset, named CompGuessWhat?!, employs a multi-task evaluation framework designed to evaluate the quality of attribute grounding during neural representation learning. It enriches the VisualGenome scene graphs linked to GuessWhat?! images, incorporating multiple tasks that predict abstract, concrete contextual, and spatial attributes. It expands the original GuessWhat?! dataset by adding additional semantic layers and attributes. Its tasks cover attribute-based language learning, which involves goal-oriented evaluation and object attribute prediction.