Originally created for Stanford’s course, this dataset is a scaled-down version of the massive ImageNet database, designed to be more manageable for training models on standard hardware while remaining complex enough for meaningful research. Content: 120,000 total images.
For Python users, this dataset is commonly loaded using libraries like or TensorFlow via torchvision.datasets or tensorflow_datasets .
200 distinct categories (e.g., animals, vehicles, everyday objects). Image Resolution: pixels (full-color JPEG format). Data Split: Training: 100,000 images (500 per class). Validation: 10,000 images (50 per class). Test: 10,000 images (unlabeled). Implementation Details COLLECTION PICS 200zip
: Organized into 200 subdirectories, each containing 500 images for that specific class.
: Includes a flat list of 10,000 images and a val_annotations.txt file that maps each image to its correct class for validation purposes. Originally created for Stanford’s course, this dataset is
When working with the tiny-imagenet-200.zip file, developers typically use a custom data loader to handle the folder structure. Below is a conceptual breakdown of the typical file organization:
: Maps those WordNet IDs to human-readable labels (e.g., "n02124075" becomes "Egyptian cat"). 200 distinct categories (e
Adding dataset Tiny-Imagenet · Issue #6127 · pytorch/vision - GitHub
