PatchCamelyon-C 和 LocalTCT-C

本研究介绍了两个用于评估深度神经网络对数字病理图像中常见腐败的鲁棒性的基准数据集：PatchCamelyon-C和LocalTCT-C。这些数据集通过在验证图像中注入九种常见腐败类型来生成，每种腐败类型有五个严重程度级别。数据集的创建旨在解决模型在面对图像腐败时的性能下降问题，特别是在数字病理图像分析领域。通过这些数据集，研究者可以评估和改进模型在实际应用中的稳定性和可靠性，尤其是在医疗诊断等关键领域。

This study presents two benchmark datasets, PatchCamelyon-C and LocalTCT-C, for assessing the robustness of deep neural networks against common corruptions in digital pathological images. Both datasets are generated by introducing nine common corruption types into validation images, with five severity levels for each corruption type. These datasets are developed to address the performance degradation of models when encountering image corruptions, especially in the domain of digital pathological image analysis. Using these datasets, researchers can evaluate and improve the stability and reliability of models in real-world applications, particularly in high-stakes fields such as medical diagnosis.