TXUXIv1, TXUXIv2, TXUXIv3

本研究引入了三个新的图像数据集：TXUXIv1、TXUXIv2和TXUXIv3，用于评估解释性人工智能（XAI）方法的忠诚度。这些数据集包含可靠的解释性真实数据，旨在通过客观比较识别忠诚度低的XAI方法，从而推动更可信和有效的XAI技术的发展。TXUXIv1以完美线条为背景，TXUXIv2使用木材纹理，TXUXIv3则包含5640种不同的背景纹理。这些数据集的创建旨在通过增加背景复杂性来挑战XAI方法，从而更全面地评估其性能。数据集的应用领域包括医疗领域等，旨在解决XAI方法的解释性和可靠性问题。

This study introduces three novel image datasets: TXUXIv1, TXUXIv2, and TXUXIv3, for evaluating the fidelity of eXplainable Artificial Intelligence (XAI) methods. These datasets contain reliable explanatory ground-truth data, aiming to identify low-fidelity XAI methods via objective comparison, thereby advancing the development of more trustworthy and effective XAI technologies. TXUXIv1 uses perfect line patterns as its background, TXUXIv2 employs wood textures, while TXUXIv3 encompasses 5,640 distinct background textures. The creation of these datasets is designed to challenge XAI methods by increasing background complexity, enabling a more comprehensive assessment of their performance. The application fields of these datasets cover healthcare and other domains, with the goal of addressing the interpretability and reliability issues of XAI methods.