This repo explains how to download & process ImageNet-1K train/val dataset for using as a dataset

• Download ImageNet-1K train/val dataset from academic torrents : train link, val link
• Check-out my velog post for download on linux server : link
• Check-out more informations on original ImageNet website : link

• ImageNet-1K train dataset zip contains zips like below

└── ILSVRC2012_img_train.tar
    ├── n01440764.tar
    ├── n01443537.tar
    ├── n01484850.tar
    ├── ...
    └── n15075141.tar

• ImageNet-1K val dataset zip contains images like below

└── ILSVRC2012_img_val.tar
    ├── ILSVRV2012_val_00000001.JPEG
    ├── ILSVRV2012_val_00000002.JPEG
    ├── ILSVRV2012_val_00000003.JPEG
    ├── ...
    └── ILSVRV2012_val_00050000.JPEG

• ImageNet_class_index.json : include class infos
  • Caution : same label with different class num exists
    • crane : 134, 517
    • maillot : 638, 639
• ImageNet_val_label.txt : include validation image label
• check.py : check if unpacked right or not
• unpack.py : make clean file trees of ILSVRC2012_img_train.tar, ILSVRC2012_img_val.tar for using as a dataset

1. Assume all the required files are in same directory like below (base_dir)

└── base_dir
    ├── ILSVRC2012_img_train.tar
    ├── ILSVRC2012_img_val.tar
    ├── ImageNet_class_index.json
    └── ImageNet_val_label.txt

2. From unpack.py, change base_dir and target_dir variables

3. Run unpack.py and it makes file trees in specific directory like below (target_dir)

└── target_dir
    ├── train
    │   ├── 0
    │   │   ├── n01440764_18.JPEG
    │   │   ├── n01440764_36.JPEG
    │   │   └── ...
    │   ├── 1
    │   ├── ...
    │   └── 999
    └── val
        ├── 0
        │   ├── ILSVRC2012_val_00000293.JPEG
        │   ├── ILSVRC2012_val_00002138.JPEG
        │   └── ...
        ├── 1
        ├── ...
        └── 999

4. From check.py, change ImageNet_dir variable and run for double-check