UzbekNet

UzbekNet is a computer vision dataset designed specifically for fine-grained food classification within traditional Uzbek cuisine. The dataset contains 4,587 images organized across 21 distinct classes. It intentionally focuses on high intra-class variance and includes regional cooking styles, plating aesthetics, and varying lighting conditions. This method provides a robust benchmark for training modern deep learning vision models, which is useful for building mobile apps that can recognize national dishes and creating personalized healthy eating plans.

Statistics

• Total images: 4587
• Granularity: 21 fine-grained classes
• Evaluation Splits: 70% Train | 15% Validation | 15% Test
• Format: Hugging Face imagefolder standard

Classes

• Achichuk: 350 images
• Buxoro Oshi Sofi: 100 images
• Chuchvara: 250 images
• Ko'k Somsa: 219 images
• Manti: 250 images
• Mastava: 150 images
• Moshkichiri: 150 images
• Non: 350 images
• Norin: 150 images
• Perashki: 250 images
• Qatlama Somsa: 200 images
• Qozon Kabob: 300 images
• Samarkand Palov: 250 images
• Shashlik: 353 images
• Sho'rva: 250 images
• Suyuq Lag'mon: 251 images
• Tandir Somsa: 150 images
• To'y Palov: 250 images
• Xonim: 110 images
• Xorazm Palov: 100 images
• Zigir Palov: 154 images

Usage

from datasets import load_dataset

ds = load_dataset("shavkatbekk/UzbekNet", split="train")

numeric_label = ds[0]["label"]

class_names = ds.features["label"].names

print(numeric_label, class_names[numeric_label])

Author

Shavkatjon Keldiyev

Citation

If you use this dataset, please link to shavkatbekk/UzbekNet.