Benchmarking Robustness of Deep Learning Classifiers Using Two-Factor Perturbation

本数据集由东南密苏里州立大学创建，旨在评估深度学习分类器在图像受到两种因素干扰下的鲁棒性。数据集包含69个测试图像集，包括一个清洁集和68个受不同程度干扰的图像集。创建过程涉及对图像应用两种类型的干扰，如盐和胡椒噪声、高斯噪声以及旋转。该数据集主要应用于无人驾驶车辆、医疗图像处理等安全关键领域，以解决深度学习分类器在实际应用中可能遇到的不完美图像处理问题。

This dataset was created by Southeast Missouri State University to evaluate the robustness of deep learning classifiers against images corrupted by two categories of disturbances. It contains 69 test image sets, consisting of one clean image set and 68 image sets corrupted to varying degrees. The dataset development process involves applying two types of disturbances to the images, namely noise-based disturbances (salt-and-pepper noise and Gaussian noise) and geometric transformations (e.g., rotation). This dataset is mainly applied in safety-critical fields such as autonomous vehicles and medical image processing, aiming to tackle the imperfect image processing issues that deep learning classifiers may encounter in practical applications.