Tiny ImageNet-200: 100,000 Images for 200 Classes, 64x64 Resolution

100,000 images across 200 classes, with 500 training, 50 validation, and 50 test images per class. This version on OpenML provides links to a 20-image subset per class for framework testing. The dataset was created by Jiayu Wu, Qixiang Zhang, and Guoxi Xu and is licensed under DbCL v1.0.

Use Cases

• Benchmarking image classification model performance based on the 200-class structure.
• Prototyping and debugging computer vision frameworks using the provided 20-image-per-class subset.
• Training and evaluating convolutional neural networks (CNNs) on a smaller-scale ImageNet variant.

Strengths

• Contains 100,000 total images, providing a substantial volume for model training.
• Offers a standardized structure with 200 classes and defined train/validation/test splits.
• The 64x64 image resolution reduces computational load for faster experimentation.

Limitations

• This specific listing only provides links to a reduced subset of 20 images per class, not the full dataset.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

Stanford University CS231n course (official website link provided).

Collection Method

Downsized and curated version of the ImageNet dataset.

Time Range

null

Freshness

Last updated date is unknown.

Geography

null