WNJXYK/TTA-CIFAR-10-C

TTA-CIFAR-10-C是CIFAR-10-C数据集的镜像版本，专门用于可重复的测试时间适应评估。该数据集包含15种不同类型的图像损坏（如高斯噪声、散粒噪声、脉冲噪声、散焦模糊、玻璃模糊、运动模糊、缩放模糊、雪、霜、雾、亮度、对比度、弹性变换、像素化和JPEG压缩），每种损坏类型有5个严重程度等级（severity_1到severity_5），每个等级包含10,000张图像。所有损坏/严重程度下都重复使用相同的10,000个CIFAR-10测试标签。该数据集主要用于评估神经网络模型在常见图像损坏和扰动下的鲁棒性。

TTA-CIFAR-10-C is a mirror of the CIFAR-10-C dataset, designed for reproducible test-time adaptation evaluation. It contains 15 different types of image corruptions (e.g., gaussian noise, shot noise, impulse noise, defocus blur, glass blur, motion blur, zoom blur, snow, frost, fog, brightness, contrast, elastic transform, pixelate, and JPEG compression), each with 5 severity levels (severity_1 through severity_5), and each level includes 10,000 images. The same 10,000 CIFAR-10 test labels are reused under every corruption/severity. This dataset is primarily used for benchmarking the robustness of neural networks to common corruptions and perturbations.