CIFAR-10 images

You can just clone this repository to use dataset. (like for timm (pytorch-image-models))

issue #2 mentioned the artifacts by jpeg compression, therefore we present a way for generating different image file extensions.

Or you prefer to generate your own image dataset, feel free to modify torch_get_pic.py. For example, if you want to generate resized images, you can add transform to CIFAR10 dataset with torchvision.

To generate images with get_pic.sh, dependencies like torch torchvision tqdm should be satisfied.

sudo chmod +x ./get_pic.sh
./get_pic.sh <1> <2> <optional: 3>

• <1>: path to your cifar-10 dataset
• <2>: image file extension: like png (no dot)
• <3>: optional: if you prefer to generate class labels in digits, then set 1 here

The script will make folders if no available output directory found.