데이터셋 탐색

Tiny ImageNet contains 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. Source: Embedded Encoder-Decoder in Convolutional Networks Towards Explainable AI Image Source: https://arxiv.org/pdf/1707.08819.pdf

Tiny ImageNetv2 is a subset of the ImageNetV2 (matched frequency) dataset by Recht et al. ("Do ImageNet Classifiers Generalize to ImageNet?") with 2,000 images spanning all 200 classes of the Tiny ImageNet dataset. It is a test set achieved by collecting images of joint classes of Tiny ImageNet and ImageNet. The resized images of size 64×64 consist of images collected from Flickr after a decade of progress on the original ImageNet dataset. The data collection process was designed to resemble the original ImageNet dataset distribution. For further information on ImageNetV2 visit the original GitHub repository of ImageNetV2.

Tiny ImageNet-R is a subset of the ImageNet-R dataset by Hendrycks et al. ("The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization") with 10,456 images spanning 62 of the 200 Tiny ImageNet dataset. It is a test set achieved by collecting images of joint classes of Tiny ImageNet and ImageNet. The resized images of size 64×64 contain art, cartoons, deviantart, graffiti, embroidery, graphics, origami, paintings, patterns, plastic objects, plush objects, sculptures, sketches, tattoos, toys, and video game renditions of ImageNet classes. For further information on ImageNet-R visit the original GitHub repository of ImageNet-R.

Tiny ImageNet-A is a subset of the Tiny ImageNet test set consisting of 3,374 images comprising real-world, unmodified, and naturally occurring examples that are misclassified by ResNet-18. The sampling process of Tiny ImageNet-A roughly follows the concept of ImageNet-A introduced by Hendrycks et al. ("Natural Adversarial Examples"). For further information on the sampling process visit the original paper.