leekwoon/oxfordflower_dataset_backup

Oxford Flower 102 Dataset Backup

Complete backup of the Oxford 102 Flower Category Dataset, a fine-grained visual classification dataset.

Download & Extract

# huggingface_hub 설치 (필요시)
pip install huggingface_hub

# 다운로드
huggingface-cli download leekwoon/oxfordflower_dataset_backup --repo-type dataset --local-dir ./oxfordflower_data

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

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

Dataset Information

Overview

The Oxford Flower 102 dataset is a consistent set of 102 flower categories commonly found in the United Kingdom. Each class consists of between 40 and 258 images. The dataset was created to address fine-grained visual classification challenges where differences between categories can be subtle.

Statistics

• Total Images: 8,189
• Training Images: 2,040 (20 per class)
• Validation Images: 1,020 (10 per class)
• Test Images: 6,149 (variable per class)
• Number of Classes: 102 flower categories
• Image Format: JPEG
• Image Size: Variable (approximately 500x500 pixels on average)

Directory Structure

oxfordflower/
├── train/                  # Training images
│   ├── 1/                 # Class 1: pink primrose
│   ├── 2/                 # Class 2: hard-leaved pocket orchid
│   ├── 3/                 # Class 3: canterbury bells
│   ├── ...
│   └── 102/               # Class 102: black-eyed susan
├── val/                    # Validation images
│   ├── 1/
│   ├── 2/
│   ├── 3/
│   ├── ...
│   └── 102/
└── test/                   # Test images
    ├── 1/
    ├── 2/
    ├── 3/
    ├── ...
    └── 102/

Flower Categories

The dataset contains 102 flower species commonly found in the UK, including:

Common Examples

• 1: pink primrose
• 2: hard-leaved pocket orchid
• 3: canterbury bells
• 4: sweet pea
• 5: english marigold
• 6: tiger lily
• 7: moon orchid
• 8: bird of paradise
• 9: monkshood
• 10: globe thistle
• ... (92 more categories)

Categories include various types of:

• Roses
• Lilies
• Orchids
• Daisies
• Tulips
• Sunflowers
• And many more

Usage

Loading with PyTorch

from torchvision import datasets, transforms

transform = transforms.Compose([
    transforms.Resize((224, 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('oxfordflower/train', transform=transform)

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

# Load test data
test_dataset = datasets.ImageFolder('oxfordflower/test', transform=transform)

Loading with TensorFlow

import tensorflow as tf

# Define preprocessing
def preprocess_image(image):
    image = tf.image.resize(image, [224, 224])
    image = tf.cast(image, tf.float32) / 255.0
    return image

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

val_ds = tf.keras.preprocessing.image_dataset_from_directory(
    'oxfordflower/val',
    image_size=(224, 224),
    batch_size=32
)

Key Challenges

This dataset presents several challenges:

• Fine-grained classification: Many flower species look very similar
• Intra-class variation: Same species can have different colors and shapes
• Scale variation: Flowers appear at different scales in images
• Background clutter: Natural images with complex backgrounds
• Pose variation: Flowers photographed from different angles

Important Notes

• Images have variable sizes and aspect ratios
• Class distribution is imbalanced in the test set
• Some images contain multiple flowers
• Background complexity varies significantly
• This is a challenging benchmark for fine-grained visual classification

Citation

If you use this dataset, please cite the original papers:

@InProceedings{Nilsback08,
  author = {Nilsback, M-E. and Zisserman, A.},
  title = {Automated Flower Classification over a Large Number of Classes},
  booktitle = {Proceedings of the Indian Conference on Computer Vision, Graphics and Image Processing},
  year = {2008}
}

@InProceedings{Nilsback06,
  author = {Nilsback, M-E. and Zisserman, A.},
  title = {A Visual Vocabulary for Flower Classification},
  booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  year = {2006},
  volume = {2},
  pages = {1447-1454}
}

License

The dataset is available for academic research purposes. For commercial use, please contact the original authors.