berkeruveyik/food_270

🍽️ Extended Food-270 Dataset

This dataset is an extended version of the original Food-101 dataset, designed to improve the diversity and representativeness of food image classification models.
It combines the well-known Food-101 dataset with additional custom food images collected and curated by Berker Üveyik to expand the number of classes and enhance the dataset’s real-world applicability.

---

📚 Overview

• Base Dataset: Food-101
• Extended Version: Food-270
• Total Classes: 270 (101 original + 169 custom classes)
• Total Images: ~ 229.245
• Split: 80% training / 20% testing
• License: CC BY-NC 4.0

---

🧠 Motivation

The original Food-101 dataset is a landmark benchmark for food recognition tasks, yet it primarily focuses on Western cuisine.
To enhance cultural and dietary diversity, this dataset extends Food-101 by adding 169 new food categories, including homemade, regional, and custom dishes.

These additional categories were carefully curated to:

• Improve the diversity and generalization capabilities of food classification models.
• Provide broader representation for non-Western and culturally specific foods.
• Enable research in nutrition tracking, food recognition, and calorie estimation applications.

This dataset is particularly suitable for:

• Food classification and computer vision research.
• Nutrition estimation and smart diet-tracking systems.
• AI-driven food recognition mobile applications.

---

🧩 Dataset Structure

Food270/
│
├── train/
│ ├── class_001/
│ │ ├── image_1.jpg
│ │ ├── image_2.jpg
│ │ └── ...
│ └── ...
└── test/
└── ...

• Train Folder: Contains 80% of the total images for training.
• Test Folder: Contains 20% of the total images for model evaluation.
• Each subfolder represents one distinct food class.

---

⚙️ Dataset Creation Process

1. Data Collection

   • The base 101 food classes were sourced directly from the Food-101 dataset.
   • 169 new classes were manually collected from diverse online sources and personal photography.

2. Data Cleaning and Curation

   • Duplicate and low-quality images were removed.
   • Labels were verified and organized consistently across classes.
   • Class distributions were balanced as much as possible.

3. Train/Test Split

   • The dataset was divided with an 80/20 ratio, ensuring that each class maintains the same proportion of samples in both sets.

---

🔐 License and Usage

This dataset is distributed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).

This means:

• ✅ You may use, share, and adapt this dataset for non-commercial purposes.
• ❌ You may not use this dataset for commercial purposes or redistribution for profit.

Why CC BY-NC 4.0?

The chosen license reflects the dataset’s educational and research-oriented purpose:

• To respect the original Food-101 creators and their open contribution.
• To encourage academic use and non-commercial AI development.
• To prevent misuse or unauthorized commercial exploitation.

---

🧾 Citation

If you use this dataset in your work, please cite both the original Food-101 dataset and this extended version (Food-270) as follows:

Original Food-101 Dataset

@inproceedings{bossard14, author = {Bossard, Lukas and Guillaumin, Matthieu and Van Gool, Luc}, title = {Food-101 – Mining Discriminative Components with Random Forests}, booktitle = {European Conference on Computer Vision (ECCV)}, year = {2014}, pages = {446--461} }

Extended Food-270 Dataset

@dataset{berkeruveyik/food_270_dataset, author = {Berker Üveyik}, title = {Food-270: An Extended Version of Food-101 with 270 Food Categories}, year = {2025}, license = {CC BY-NC 4.0}, note = {Based on and extended from the original Food-101 dataset (Bossard et al., ECCV 2014)} }

Extended Food-270 Dataset

@dataset{berkeruveyik/food_270_dataset, author = {Berker Üveyik}, title = {Food-270: An Extended Version of Food-101 with 270 Food Categories}, year = {2025}, license = {CC BY-NC 4.0}, note = {Based on and extended from the original Food-101 dataset (Bossard et al., ECCV 2014)} }

---

🙏 Acknowledgements

• The original Food-101 dataset:
  Bossard, Lukas, Matthieu Guillaumin, and Luc Van Gool. "Food-101 – Mining Discriminative Components with Random Forests." ECCV 2014.

• Special thanks to the Hugging Face community for promoting open and responsible dataset sharing.

---

👨‍💻 Author

Berker Üveyik
Software Engineer | Machine Learning Researcher
📧 Contact: [berkeruveyik@gmail.com]
🌐 https://huggingface.co/berkeruveyik

---

📚 Acknowledgement

This dataset is based on the Food-101 dataset created by Lukas Bossard, Matthieu Guillaumin, and Luc Van Gool.

If you use this dataset, please cite both the original Food-101 paper and this extended Food-270 version as follows:

Original Food-101 Dataset

Bossard, L., Guillaumin, M., & Van Gool, L. (2014).
Food-101 – Mining Discriminative Components with Random Forests.
European Conference on Computer Vision (ECCV), 446–461.
https://www.vision.ee.ethz.ch/datasets_extra/food-101/

This dataset was created for educational and research purposes to advance the field of food recognition and nutrition tracking.