leekwoon/imagenet_dataset_backup

ImageNet 2012 Dataset Backup

Complete backup of the ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2012 dataset.

Download & Extract

# huggingface_hub 설치 (필요시)
pip install huggingface_hub

# 다운로드
huggingface-cli download leekwoon/imagenet_dataset_backup --repo-type dataset --local-dir ./imagenet_data

# 무결성 확인 (선택사항)
cd imagenet_data
md5sum -c checksums.md5

# 파일 합치기 및 압축 해제
cat data.tar.gz.part_* | tar -xzvf -

Dataset Information

Overview

ImageNet is a large-scale hierarchical image database that has been instrumental in advancing deep learning and computer vision research. The ILSVRC 2012 subset is the most commonly used benchmark for image classification tasks.

Statistics

• Training Images: 1,281,167
• Validation Images: 50,000
• Test Images: 100,000
• Number of Classes: 1,000
• Image Format: JPEG
• Average Resolution: ~469x387 pixels

Directory Structure

imagenet2012/
├── train/                  # Training images
│   ├── n01440764/         # Class folder (tench)
│   ├── n01443537/         # Class folder (goldfish)
│   ├── n01484850/         # Class folder (great white shark)
│   └── ...                # 997 more class folders
├── val/                    # Validation images
│   ├── n01440764/
│   ├── n01443537/
│   ├── n01484850/
│   └── ...
└── test/                   # Test images (if available)
    └── ...

Class Information

The dataset contains 1,000 object classes from the WordNet hierarchy, including:

• Animals (mammals, birds, fish, reptiles, etc.)
• Plants (trees, flowers, fruits, vegetables)
• Objects (vehicles, furniture, tools, instruments)
• Scenes and structures

Each class is identified by a WordNet ID (synset), such as:

• n01440764: tench (a type of fish)
• n02119789: kit fox
• n07734744: mushroom
• n04515003: upright piano

Usage

Loading with PyTorch

from torchvision import datasets, transforms

transform = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406],
                       std=[0.229, 0.224, 0.225]),
])

# Load training data
train_dataset = datasets.ImageFolder('imagenet2012/train', transform=transform)

# Load validation data
val_dataset = datasets.ImageFolder('imagenet2012/val', transform=transform)

Loading with TensorFlow

import tensorflow as tf

def preprocess_image(image):
    image = tf.image.resize(image, [224, 224])
    image = tf.keras.applications.imagenet_utils.preprocess_input(image)
    return image

# Load dataset
train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    'imagenet2012/train',
    image_size=(224, 224),
    batch_size=32
)

Important Notes

• This dataset is for research purposes only
• The original ImageNet dataset requires accepting the terms of use
• Some images may be missing due to broken URLs in the original dataset
• Class labels follow the ILSVRC 2012 convention

Citation

If you use this dataset, please cite the original ImageNet paper:

@article{deng2009imagenet,
  title={ImageNet: A large-scale hierarchical image database},
  author={Deng, Jia and Dong, Wei and Socher, Richard and Li, Li-Jia and Li, Kai and Fei-Fei, Li},
  journal={2009 IEEE Conference on Computer Vision and Pattern Recognition},
  pages={248--255},
  year={2009},
  organization={IEEE}
}

License

The annotations in this dataset are licensed under a Creative Commons Attribution 4.0 International License. The images have various licenses that should be checked on the original ImageNet website.