WNJXYK/TTA-ImageNet-C

TTA-ImageNet-C是ImageNet-C的一个镜像版本，专门用于可重复的测试时间适应评估。该数据集包含15种不同的图像损坏类型（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），每种类型有5个严重程度的分割（severity_1到severity_5）。每个分割包含50,000张图像，覆盖1000个类别。数据集的标签是ClassLabel类型，包含1000个WordNet-ID名称，按torchvision顺序排列。该数据集主要用于评估神经网络在常见损坏和扰动下的鲁棒性，遵循CC BY 4.0许可协议。TTA-ImageNet-C is a mirror of ImageNet-C designed for reproducible test-time adaptation evaluation. The dataset includes 15 different corruption types (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 splits (severity_1 through severity_5). Each split contains 50,000 images covering 1000 classes. The labels are ClassLabel with 1000 WordNet-ID names in torchvision order (lexicographic on wnid). The dataset is primarily used for benchmarking neural network robustness to common corruptions and perturbations and is licensed under CC BY 4.0.