Table 2.

A comprehensive overview of notable publications for 4 volume estimation approaches, arranged chronologically.

Approach and study, year		Projects or team	Reference object	Item	Reported error
Pixel density approach
	Martin et al [13], 2009	—a	Physical card	N/Ab	N/A
	Jia et al [63], 2012	University of Pittsburgh	Circular plate Circular LED light	—	<27.60 <54.10
	Pouladzadeh et al [32], 2014	—	User’s thumb	5	<10
	Okamoto and Yanai [64], 2016	UECc	Wallet	3	Mean calorie error Beef rice bowl –242 (SD 55.1) Croquette –47.08 (SD 52.5) Salad 4.86 (SD 11.9)
	Akpa et al [65], 2017	—	Chopstick	15	<6.65
	Liang and Li [66], 2017	—	1-yuan coin	19 fruits	15 items <20%
	Yanai et al [67], 2019 and Ege et al [67], 2019	UEC	Rice grain size	3	<10%
Geometric modeling approach
	Zhu et al [24], 2010 and Zhu et al [25], 2008	TADAd	Checkerboard	7	Spherical 5.65% Prismatic 28.85%
	Chae et al [69], 2011	TADA	Checkerboard	26	Cylinders 11.1% Flattop solid 11.7%
	Chen et al [70], 2013	University of Pittsburgh	Circular plate	17	3.69%
	Jia et al [71], 2014	University of Pittsburgh	Circular plate Other container	100	<30% from 85/100 of test items
	Tanno et al [72], 2018	UEC	Apple ARKit	3	Mean calorie error Beef rice bowl –67.14 (SD 18.8) Croquette–127.0 (SD 9.0) Salad –0.95 (SD 0.16)
	Yang et al [73], 2019	University of Pittsburgh	Augmented reality	15	Large objects 16.65% Small objects 47.60%
	Smith et al [74], 2022	—	Checkerboard	26	Single food items 32.4%-56.1% Multiple food items 23.7%-32.6%
3D reconstruction approach
	Puri et al [26], 2009	—	3 images Checkerboard	26	2%-9.5%
	Kong and tan [75], 2012	—	3 images Checkerboard	7	Volume estimation error 20%
	Rahman et al [76], 2012	TADA	2 images Checkerboard	6	7.70%
	Chang et al [77], 2013	TADA	Using food silhouettes to reconstruct a 3D object	4	10%
	Anthimopoulos et al [78], 2015	GoCARB	2 images physical card Physical card	N/A	Volume estimation error 9.4%
	Dehais et al [79], 2017	GoCARB	2 images Physical card	45 dishes 14 meals	8.2%-9.8%
	Gao et al [80], 2018	—	SLAMe-based with Rubik cube	3	11.69%-19.20% for static measurement 16.32%-27.9% for continuous measurement
	Ando et al [81], 2019	UEC	Multiple cameras on iPhone X for depth estimation	3	Calorie estimation error Sweet and sour pork <1% Fried chicken <1% Croquette <15%
	Lu et al [58], 2020	GoCARB	2 images Physical card and gravity information	234 items from MADiMAf	MAREg 19%, while their earlier system, GoCarb (2017), achieved 22.6% on the same task [79].
Depth camera approach
	Shang et al [82], 2011	—	Specific food recording device	—	No performance report
	Chen et al [83], 2012	—	Depth camera	—	No performance report
	Fang et al [84], 2016	TADA	Camera from this study [85]	10	Depth method overestimates volume than geometric model
	Alfonsi et al [86], 2020	—	iPhone and Android devices	200	Carbohydrate estimation error <10 g
	Herzig et al [87], 2020		iPhone X	128	Relative error of weight estimation 14.0%

^aNot available.

^bN/A: not applicable.

^cUEC: University of Electro-Communications.

^dTADA: Technology Assisted Dietary Assessment.

^eSLAM: simultaneous localization and mapping.

^fMADiMA: Multimedia Assisted Dietary Management.

^gMARE: mean absolute relative error.