ImageNet16 Dataset

Overview

ImageNet16 is a compact subset of the ImageNet dataset designed for faster experimentation and research with limited computational resources. The dataset contains 16 object categories selected from ImageNet-1K and is suitable for image classification tasks.

Statistics

Split	Images
Train	6,400
Test	1,600
Total	8,000

Each class contains:

• 400 training images
• 100 test images

Classes

• Bird (American Egret)
• Dog (Toy Poodle)
• Cat (Siamese Cat)
• Bear (Brown Bear)
• Elephant (African Elephant)
• Airplane (Airliner)
• Bicycle
• Boat (Canoe)
• Knife (Cleaver)
• Keyboard
• Clock
• Truck (Police Van)
• Chair (Rocking Chair)
• Oven (Rotisserie)
• Car (Sports Car)
• Bottle (Wine Bottle)

Dataset Structure

train/
├── class_1/
├── class_2/
└── ...

val/
├── class_1/
├── class_2/
└── ...

Images are stored in JPEG format.

Applications

• Image Classification
• Lightweight Deep Learning Research
• Neural Architecture Search
• Efficient CNN Evaluation
• Edge AI and Embedded Vision

Source

Original Zenodo record:

https://zenodo.org/records/8027520

Citation

@article{kyrkou2024structured,
  title={Toward Efficient Convolutional Neural Networks With Structured Ternary Patterns},
  author={Kyrkou, Christos},
  journal={IEEE Transactions on Neural Networks and Learning Systems},
  year={2024},
  doi={10.1109/TNNLS.2024.3380827}
}

Acknowledgements

This dataset is derived from the original ImageNet dataset. Credit goes to the original ImageNet creators:

@article{russakovsky2015imagenet,
  title={ImageNet Large Scale Visual Recognition Challenge},
  author={Russakovsky, Olga and Deng, Jia and Su, Hao and others},
  journal={International Journal of Computer Vision},
  volume={115},
  number={3},
  pages={211--252},
  year={2015}
}