LONG-TAILED FOOD CLASSIFICATION Jiangpeng HeLuotao LinyHeather Eicher-MilleryFengqing Zhu School of Electrical and Computer EngineeringyDepartment of Nutrition Science

2025-05-02 0 0 777.28KB 5 页 10玖币

侵权投诉

LONG-TAILED FOOD CLASSIFICATION

Jiangpeng He∗Luotao Lin†Heather Eicher-Miller†Fengqing Zhu∗

∗School of Electrical and Computer Engineering †Department of Nutrition Science

Purdue University, West Lafayette, Indiana, U.S.A

ABSTRACT

Food classiﬁcation serves as the basic step of image-based

dietary assessment to predict the types of foods in each input

image. However, food image predictions in a real world sce-

nario are usually long-tail distributed among different food

classes, which cause heavy class-imbalance problems and a

restricted performance. In addition, none of the existing long-

tailed classiﬁcation methods focus on food data, which can

be more challenging due to the lower inter-class and higher

intra-class similarity among foods. In this work, we ﬁrst

introduce two new benchmark datasets for long-tailed food

classiﬁcation including Food101-LT and VFN-LT where the

number of samples in VFN-LT exhibits the real world long-

tailed food distribution. Then we propose a novel 2-Phase

framework to address the problem of class-imbalance by (1)

undersampling the head classes to remove redundant sam-

ples along with maintaining the learned information through

knowledge distillation, and (2) oversampling the tail classes

by performing visual-aware data augmentation. We show

the effectiveness of our method by comparing with existing

state-of-the-art long-tailed classiﬁcation methods and show

improved performance on both Food101-LT and VFN-LT

benchmarks. The results demonstrate the potential to apply

our method to related real life applications.

Index Terms—Image Classiﬁcation, Long-tailed Distri-

bution, Deep Learning, Knowledge Distillation

1. INTRODUCTION

Accurate identiﬁcation of food is critical to image-based di-

etary assessment [1, 2], which facilitates matching the food

to the proper identiﬁcation of that food in a nutrient database

with corresponding nutrient composition [3]. Such linkage

makes it possible to determine dietary links to health and dis-

ease, such as diabetes. Dietary assessment, therefore, is very

important to health-care related applications such as [4, 5] due

to recent advances in novel computation approaches and new

sensor devices. The performance of image-based dietary as-

sessment relies on the accurate prediction of foods in the cap-

tured eating scene images. However, real world food images

usually have a long-tailed distribution where a small portion

Fig. 1. The overview of the VFN-LT that exhibits real-world

long-tailed food distribution. The number of training sam-

ples are assigned based on consumption frequency, which is

matched through NHANES from 2009–2016 among 17,796

U.S. healthy adults.

of food classes (i.e. head class) contain abundant samples for

training while most food classes (i.e. tail class) have only a

few samples as shown in Figure 1. The long-tailed classiﬁca-

tion, deﬁned as the extreme class-imbalance problem, leads

to classiﬁcation bias towards head classes and poor general-

ization ability on recognizing tail food classes.

As few existing long-tailed image classiﬁcation methods

target food images, we ﬁrst introduce two benchmark long-

tailed food datasets including Food101-LT and VFN-FT. Sim-

ilarly as in [6], Food101-LT is constructed as a long-tailed

version of the original balanced Food101 [7] dataset by fol-

lowing the Pareto distribution. In addition, as shown in Fig-

ure 1, VFN-LT is also used and provides a new and valuable

long-tailed distributed food dataset where the number of sam-

ples for each food class exhibits the distribution of consump-

tion frequency [8], deﬁned as how often a food is consumed

in one day as in the National Health and Nutrition Exami-

arXiv:2210.14748v1 [cs.CV] 26 Oct 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

LONG-TAILEDFOODCLASSIFICATIONJiangpengHeLuotaoLinyHeatherEicher-MilleryFengqingZhuSchoolofElectricalandComputerEngineeringyDepartmentofNutritionSciencePurdueUniversity,WestLafayette,Indiana,U.S.AABSTRACTFoodclassicationservesasthebasicstepofimage-baseddietaryassessmenttopredictthetypesoffoodsine...

展开>> 收起<<

LONG-TAILED FOOD CLASSIFICATION Jiangpeng HeLuotao LinyHeather Eicher-MilleryFengqing Zhu School of Electrical and Computer EngineeringyDepartment of Nutrition Science.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

LONG-TAILED FOOD CLASSIFICATION Jiangpeng HeLuotao LinyHeather Eicher-MilleryFengqing Zhu School of Electrical and Computer EngineeringyDepartment of Nutrition Science

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: