PyTorch built-in datasets

• CIFAR-10: 60,000 labeled 32x32 color images of 10 classes of objects
• CIFAR-100: 60,000 labeled 32x32 color images of 100 fine classes of objects, also grouped into 20 course superclasses
• EMNIST: 800,000 labeled 28x28 grayscale images of handwritten digits and letters (uppercase and lowercase) that expand the MNIST dataset
• Fashion-MNIST: 70,000 labeled 28x28 grayscale images of 10 classes of clothing articles
• ImageNet 2012: 1,331,167 labeled color images of tens of thousands of classes of nouns in the WordNet hierarchy
• KMNIST: 70,000 labeled 28x28 grayscale images of 10 classes of handwritten Japanese Hiragana characters
• LSUN: 708,564 labeled large-scale color images of 10 classes of scenes/settings
• MNIST: 70,000 labeled 28x28 grayscale images of 10 classes of handwritten digits
• SVHN: Color images of house numbers (addresses) obtained from Google Street View w/ labeled bounding boxes around individual digits

TensorFlow built-in datasets

• Cats and Dogs: 25,000 labeled images of cats and dogs
• CelebA: 202,599 total face images of 10,177 identities w/ facial landmarks, aligned & cropped images, bounding boxes, and binary attribute labels
• CIFAR-10: 60,000 labeled 32x32 color images of 10 classes of objects
• CIFAR-100: 60,000 labeled 32x32 color images of 100 fine classes, also grouped into 20 course superclasses
• CIFAR-10-C: Images from CIFAR-10 manipulated using 15 common corruptions at 5 levels of severity each
• Colorectal Histology: 5,000 labeled 150x150 color histological images of 8 tissue types of human colorectal cancer
• CBIS-DDSM: 2,620 labeled scanned film mammography studies of normal, benign, and malignant cases of human breast cancer
• Diabetic Retinopathy Detection: 88,712 labeled images of eyes with diabetic retinopathy at none, mild, moderate, severe, and proliferative levels of severity
• EMNIST: 800,000 labeled 28x28 grayscale images of handwritten digits and letters (uppercase and lowercase) that expand the MNIST dataset