ModelTC / Imagenet-S

Robustness for real-world system noise

ImageNet-S

Robustness for real-world system noise

Usage 1 --> Generate dataset on file system

from imagenet_s_gen import ImageTransfer
ImageTransfer(root_dir='/your/val/images/root/path', meta_file='/meta/val.txt',
              save_root='/your/save/root/path', decoder_type='pil',
              transform_type='val', resize_type='pil-bilinear').write_to_filesystem()

Usage 2 --> Generate one image real time (recommend)

from imagenet_s_gen import ImageTransfer
image_gen = ImageTransfer(root_dir='/your/val/images/root/path', meta_file='/meta/val.txt',
            save_root='', decoder_type='pil',
            transform_type='val', resize_type='pil-bilinear')
# generate numpy image on index 0
numpy_image, label = image_gen.getimage(0)