Datasets:

---

ILSVRC

/

imagenet-1k

like 611

Follow

Large Scale Visual Recognition Challenge 65

Extract with progress and reorganize into train/<class> and val/<class>.

#!/usr/bin/env bash
set -e
mkdir -p ../train ../val
for f in train_images_*.tar.gz; do
  echo "Extracting $f → ../train"
  pv "$f" | tar -xz -C ../train
done
echo "Extracting val_images.tar.gz → ../val"
pv val_images.tar.gz | tar -xz -C ../val
echo "Done."

#!/usr/bin/env python3
import os, shutil, argparse
from tqdm import tqdm
from classes import IMAGENET2012_CLASSES

def organize_split(split_dir, class_ids):
    for cls in class_ids:
        os.makedirs(os.path.join(split_dir, cls), exist_ok=True)
    files = [f for f in os.listdir(split_dir) if os.path.isfile(os.path.join(split_dir, f))]
    for fname in tqdm(files, desc=f"Organizing {os.path.basename(split_dir)}", unit="file"):
        ext = os.path.splitext(fname)[1]
        matched = next((cls for cls in class_ids
                        if fname.startswith(f"{cls}_") or fname.endswith(f"_{cls}{ext}")), None)
        if not matched:
            tqdm.write(f"[WARN] skipping {fname}")
            continue
        shutil.move(os.path.join(split_dir, fname),
                    os.path.join(split_dir, matched, fname))

def main():
    p = argparse.ArgumentParser()
    p.add_argument("-d", "--data-dir", required=True)
    args = p.parse_args()
    class_ids = list(IMAGENET2012_CLASSES.keys())
    for split in ("train", "val"):
        organize_split(os.path.join(args.data_dir, split), class_ids)

if __name__ == "__main__":
    main()

Usage:

./extract.sh    # from imagenet/data
python organize_imagenet.py --data-dir ../imagenet