shuqike/tiny-imagenet-200-clean

Dataset Card for tiny-imagenet-200-clean

Dataset Description

• —Homepage: https://www.kaggle.com/c/tiny-imagenet
• —Repository: [Needs More Information]
• —Paper: http://cs231n.stanford.edu/reports/2017/pdfs/930.pdf
• —Leaderboard: https://paperswithcode.com/sota/image-classification-on-tiny-imagenet-1

Dataset Summary

The original Tiny ImageNet contained 100000 images of 200 classes (500 for each class) downsized to 64×64 colored images. Each class has 500 training images, 50 validation images, and 50 test images.

This clean version removed grey scale images and only kept RGB images.

Languages

The class labels in the dataset are in English.

Dataset Structure

Data Instances

{
  'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=64x64 at 0x1A800E8E190,
  'label': 15
}

Data Fields

• —image: A PIL.Image.Image object containing the image.
• —label: an int classification label.

Data Splits

Usage

Example

Load Dataset

def example_usage():
    tiny_imagenet = load_dataset('slegroux/tiny-imagenet-200-clean', split='train')
    print(tiny_imagenet[0])

if __name__ == '__main__':
    example_usage()