Central Asian Food Dataset Card

Repository: https://github.com/IS2AI/Kazakh-Food-Dataset

Summary:

The Central Asian Food Dataset (CAFD) is a new dataset containing 16,499 images of food items across 42 classes. This dataset addresses the lack of publicly available resources for food image recognition in Central Asia. The dataset is imbalanced, with class distribution shown in the provided figures. Several pre-trained models (VGG-16, Squeezenet1_0, ResNet50, ResNet101, ResNet152, ResNext50_32, Wide ResNet-50, DenseNet-121, and EfficientNet-b4) have been trained on this dataset and its performance evaluated against a larger food dataset (Food1K). Pre-trained model weights are available for download. The dataset and pre-trained models are useful for developing personalized dietary intervention systems and food image recognition applications specific to Central Asian cuisine.

Dataset Statistics:

Number of Images: 16,499
Number of Classes: 42
Image Samples: (See image below)
Class Distribution: (See image below)

Model Performance:

The table below presents the Top-1 and Top-5 accuracy results for various models trained on CAFD and a combination of CAFD and the Food1K dataset.

Model	CAFD (Top-1 Acc.)	CAFD (Top-5 Acc.)	Food1K+CAFD (Top-1 Acc.)	Food1K+CAFD (Top-5 Acc.)
VGG-16	86.03	98.33	80.87	96.19
Squeezenet1_0	79.58	97.29	69.16	90.15
ResNet50	88.03	98.44	83.22	97.25
ResNet101	88.51	98.44	84.20	97.45
ResNet152	88.70	98.59	84.75	97.58
ResNext50_32	87.95	98.44	84.81	97.65
Wide ResNet-50	88.21	98.59	85.27	97.81
DenseNet-121	86.95	98.26	82.45	96.93
EfficientNet-b4	81.28	97.37	87.75	98.01

Citations:

[1] Min, Weiqing and Wang, Zhiling (2021). Large Scale Visual Food Recognition. arXiv.

@Article{nu15071728,
AUTHOR = {Karabay, Aknur and Bolatov, Arman and Varol, Huseyin Atakan and Chan, Mei-Yen},
TITLE = {A Central Asian Food Dataset for Personalized Dietary Interventions},
JOURNAL = {Nutrients},
VOLUME = {15},
YEAR = {2023},
NUMBER = {7},
ARTICLE-NUMBER = {1728},
URL = {https://www.mdpi.com/2072-6643/15/7/1728},
ISSN = {2072-6643}
}