ChineseFoodNet: A large-scale Image Dataset for Chinese Food Recognition

TL;DR

ChineseFoodNet is a large-scale, diverse image dataset for Chinese food recognition, featuring over 180,000 images across 208 categories, with a novel fusion method improving CNN accuracy.

Contribution

The paper introduces ChineseFoodNet, a comprehensive dataset with real dish images, and proposes TastyNet, a novel CNN fusion approach for improved recognition accuracy.

Findings

Achieved over 81% top-1 accuracy with TastyNet.

Dataset includes diverse real-world Chinese food images.

Demonstrated effectiveness of machine learning in reducing labeling effort.

Abstract

In this paper, we introduce a new and challenging large-scale food image dataset called "ChineseFoodNet", which aims to automatically recognizing pictured Chinese dishes. Most of the existing food image datasets collected food images either from recipe pictures or selfie. In our dataset, images of each food category of our dataset consists of not only web recipe and menu pictures but photos taken from real dishes, recipe and menu as well. ChineseFoodNet contains over 180,000 food photos of 208 categories, with each category covering a large variations in presentations of same Chinese food. We present our efforts to build this large-scale image dataset, including food category selection, data collection, and data clean and label, in particular how to use machine learning methods to reduce manual labeling work that is an expensive process. We share a detailed benchmark of several state-of-the-art deep convolutional neural networks (CNNs) on ChineseFoodNet. We further propose a novel two-step data fusion approach referred as "TastyNet", which combines prediction results from different CNNs with voting method. Our proposed approach achieves top-1 accuracies of 81.43% on the validation set and 81.55% on the test set, respectively. The latest dataset is public available for research and can be achieved at https://sites.google.com/view/chinesefoodnet.