Processing tiny-imagenet-200 datasets to pytorch.dataset.ImageFolders

build_train_data.py

# the file is at: data/tiny-imagenet-200/train/
import os
dirs = os.listdir('.')  # get categoris
for the_dir in dirs:
    if os.path.isdir(the_dir):  # only work for folders
        for _, _, images in os.walk(os.path.join(the_dir, 'images')):
            for image in images:
                os.rename(
                    os.path.join(the_dir, 'images', image),
                    os.path.join(the_dir, image)
                )  # extract the images from 'images' folder to category root
        os.rmdir(os.path.join(the_dir, 'images'))  # remove 'images' folder
        os.remove(os.path.join(the_dir, the_dir + '_boxes.txt'))  # remove the boxes file

build_val_data.py

# the file is at: data/tiny-imagenet-200/val/
import os
sheet_path = 'val_annotations.txt'  # reference
with open(sheet_path, 'r') as f:
    lines = f.readlines()  # read the file
for line in lines:
    filename, catogory = line.split('\t')[:2]  # the first two words from each line
    if os.path.isdir(catogory) == False:
        os.mkdir(catogory)  # make dir in advance
    os.rename('images/' + filename, catogory + '/' + filename)