tiny-imagenet-200: 100,000 Images for 200-Class Object Recognition

100,000 images across 200 classes, with 500 training, 50 validation, and 50 test images per class. The dataset consists of 64x64 colored images and was created by Jiayu Wu, Qixiang Zhang, and Guoxi Xu for a Stanford CS231n project. It is licensed under DbCL v1.0.

Use Cases

• Benchmarking image classification models based on the 200-class structure.
• Training lightweight neural networks based on the 64x64 resolution images.
• Comparing model performance on validation and test splits as described in the dataset structure.

Strengths

• Contains 100,000 images, providing a substantial scale for model training.
• Structured with 500 training, 50 validation, and 50 test images per class, enabling standardized evaluation.
• Images are consistently downsized to 64x64 resolution, simplifying preprocessing.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• The dataset here just contains links to the images, requiring a separate download to access the actual files.

Provenance

Source

Stanford CS231n course project.

Collection Method

Images were downsized from the original ImageNet dataset.

Time Range

null

Freshness

Last updated is unknown.

Geography

null