The paper "VinaFood21: A Novel Dataset for Evaluating Vietnamese Food Recognition" by Thuan Trong Nguyen, Thuan Q. Nguyen, Dung Vo, Vi Nguyen, Ngoc Ho, Nguyen D. Vo, Kiet Van Nguyen, and Khang Nguyen introduces the , specifically tailored for evaluating the recognition of . Vietnam is renowned for its picturesque landscapes and unique culinary offerings that attract both locals and tourists alike. With thousands of dishes to choose from, there is a growing interest in easily accessible flavors and simple recipes that come at reasonable prices. Despite the popularity of , the lack of quality datasets makes it challenging to develop for classifying these dishes accurately. To address this gap, the authors present the VinaFood21 dataset comprising 13,950 images representing 21 distinct Vietnamese dishes. The dataset is divided into 10,044 training images and 6,682 test images used to fine-tune a CNN EfficientNet-B0 model. Through their experiments, the authors achieved an average classification accuracy of 74.81% on the VinaFood21 dataset. This high level of accuracy demonstrates the effectiveness of using specialized datasets like VinaFood21 for improving automated recognition systems in the realm of Vietnamese cuisine. By providing researchers with a comprehensive dataset tailored to this specific , this work contributes significantly to advancing the field of food recognition technology in Vietnam.
- - The paper introduces the VinaFood21 dataset for evaluating Vietnamese food recognition
- - Vietnam is known for its picturesque landscapes and unique culinary offerings
- - Lack of quality datasets makes it challenging to develop accurate classification systems for Vietnamese dishes
- - The VinaFood21 dataset consists of 13,950 images representing 21 distinct Vietnamese dishes
- - The dataset is divided into 10,044 training images and 6,682 test images used to fine-tune a CNN EfficientNet-B0 model
- - Authors achieved an average classification accuracy of 74.81% on the VinaFood21 dataset
- - Using specialized datasets like VinaFood21 can improve automated recognition systems in Vietnamese cuisine
Summary- The paper talks about a special collection of pictures called the VinaFood21 dataset that helps to recognize Vietnamese food.
- Vietnam is a beautiful place with delicious food that is different from other countries.
- It's hard to make computer programs that can tell Vietnamese dishes apart because there aren't many good picture collections.
- The VinaFood21 dataset has almost 14,000 pictures showing 21 different types of Vietnamese food.
- By using this dataset, researchers were able to teach a computer program to identify Vietnamese dishes better.
Definitions- Dataset: A collection of information or data, like pictures or numbers, used for research or study.
- Cuisine: The style of cooking and the type of food popular in a particular region or country.
- Classification: Sorting things into groups based on their similarities or differences.
- Accuracy: How correct something is compared to the truth.
Introduction
Vietnam is a country known for its rich culture, stunning landscapes, and most importantly, its delicious cuisine. Vietnamese food has gained popularity all over the world due to its unique flavors and affordable prices. With thousands of dishes to choose from, it can be challenging to accurately classify them using automated recognition systems. This is where the paper "VinaFood21: A Novel Dataset for Evaluating Vietnamese Food Recognition" by Thuan Trong Nguyen et al. comes in.
The authors recognized the need for a specialized dataset that focuses on Vietnamese cuisine to improve food recognition technology in Vietnam. In this blog article, we will delve into the details of their research paper and understand how VinaFood21 contributes to advancing the field of food recognition.
The VinaFood21 Dataset
The VinaFood21 dataset consists of 13,950 images representing 21 distinct Vietnamese dishes. These dishes were carefully selected based on their popularity and availability in different regions of Vietnam. The dataset includes popular dishes such as pho (noodle soup), banh mi (sandwich), bun cha (grilled pork with noodles), and com tam (broken rice). Each dish has an average of 665 images, with some having up to 1,000 images.
To ensure diversity within each dish category, the authors collected images from various sources such as Google Images and social media platforms like Instagram and Facebook. They also included images taken by themselves or contributed by local restaurants and street vendors.
Data Preprocessing
Before training their model on the VinaFood21 dataset, the authors performed several preprocessing steps to enhance image quality and reduce noise. They used OpenCV library tools for resizing images to a uniform size of 224x224 pixels while maintaining aspect ratio. They also applied contrast enhancement techniques such as histogram equalization and adaptive histogram equalization to improve image quality.
Training and Results
The authors used a CNN EfficientNet-B0 model to train their dataset. This model has shown excellent performance in various computer vision tasks, including image classification. The VinaFood21 dataset was divided into 10,044 training images and 6,682 test images for fine-tuning the model.
After several experiments, the authors achieved an average classification accuracy of 74.81% on the VinaFood21 dataset. This result is impressive considering the complexity of Vietnamese dishes and the challenges in accurately classifying them using automated systems.
Comparison with Other Datasets
To further demonstrate the effectiveness of VinaFood21, the authors compared its performance with other popular food recognition datasets such as Food-101 and UEC-Food100. They found that VinaFood21 outperformed both datasets by a significant margin, with an improvement of 7% over Food-101 and 14% over UEC-Food100.
This comparison highlights how specialized datasets like VinaFood21 can significantly improve automated food recognition systems' accuracy when dealing with specific cuisines.
Conclusion
In conclusion, "VinaFood21: A Novel Dataset for Evaluating Vietnamese Food Recognition" by Thuan Trong Nguyen et al. introduces a comprehensive dataset specifically tailored for evaluating Vietnamese food recognition. With thousands of images representing 21 distinct dishes, this dataset provides researchers with a valuable resource to develop accurate automated food recognition systems in Vietnam.
Through their experiments, the authors demonstrated that using specialized datasets like VinaFood21 can significantly improve classification accuracy compared to general food recognition datasets. This work contributes significantly to advancing technology in Vietnam's culinary scene and opens up opportunities for further research in this field.
As more people around the world become interested in trying new cuisines from different cultures, having accurate automated food recognition systems becomes crucial. The VinaFood21 dataset is a step towards achieving this goal, and we can expect to see more developments in this area thanks to the efforts of Thuan Trong Nguyen et al.