Kidsize is a proof-of-concept developed by IBV. This presentation describes the two innovations underpinning Kidsize concept and presents the results of their validation. The first one is a mobile phone app to measure a child in 3D by taking two pictures. This new method is more accurate and consistent than an untrained person using a measuring tape at home or in the shop. The second one is an expert system that recommends the size that best fits the child and assesses the fit of the garment at different body areas. Project results show that it can provide nearly 90% right size recommendations, thus outperforming existing methods like age- or height-based size guides, which achieve 40 and 60% right recommendations respectively.
Boost PC performance: How more available memory can improve productivity
Kidsize: always get the right size! @3DBody.Tech 1st Dec 2016
1. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Always get the right size!
Alfredo Ballester & Ana Piérola
2. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
IBV is a private not-for-profit R&D organisation
Product design consultancy
for manufacturing industries
Technology development partner
for IT companies and Academia
Apparel Sports Transport
Health
Safety
Leisure
Appliances Elderly
Orthotics
Conduction of 3D body scanning surveys
Data-driven 3D modelling from scans (3D)
images (2D) or body measurements (1D)
Data analytics & services
(measuring, shape analyses, size advice)
Applications & hardware prototyping
3. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Introduction
Body measuring app
Size advice algorithms
Validation & results
Conclusions
Ongoing work
4. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Introduction
Body measuring app
Size advice algorithms
Validation & results
Conclusions
Ongoing work
5. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Why kidsize?
Ingredients for size advice
6. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Proof-of-concept for Inditex
Why kidsize?
Brand exp. 41% 21% 64% / 55%
Size guide 41% 5% 52% / 43%
IBV model 86% 93% 83% / 93%
SIZE SELECTION
MODEL
Fitting
percention on
body parts
Size selection
FITTING MOLDEL
OF BODY PARTS
FITTING
PROBABILITY OF
KEY AREAS
CONTRIBUTION OF
FITTING AREAS
STAGE 2
SIZE RECOMENDED
STAGE 1
FITTING PREDICTION
ON KEY BODY PARTS
Body measurements Garment sizes
7. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Proof-of-concept for Inditex
Why kidsize?
Brand exp. 41% 21% 64% / 55%
Size guide 41% 5% 52% / 43%
IBV model 86% 93% 83% / 93%
SIZE SELECTION
MODEL
Fitting
percention on
body parts
Size selection
FITTING MOLDEL
OF BODY PARTS
FITTING
PROBABILITY OF
KEY AREAS
CONTRIBUTION OF
FITTING AREAS
STAGE 2
SIZE RECOMENDED
STAGE 1
FITTING PREDICTION
ON KEY BODY PARTS
Body measurements Garment sizes
• Bodies measured with Vitus XXL scanner
• All garments and sizes measured by hand
• Single garment models trained with fit trials
8. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Why kidsize?
Co-funded by the EC (2013-2016)
Grant Agreement no. 606091
Industry Promoters DemonstratorsDevelopers
9. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Key developments
body
measuring
instrument
10. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Key developments
body
measuring
instrument
Size advice &
fit prediction
algorithms
11. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Key developments
body
measuring
instrument
Size advice &
fit prediction
algorithms
12. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
How does it work?
13. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Introduction
Body measuring app
Size advice algorithms
Validation & results
Conclusions
Ongoing work
14. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Body measuring app
measured in
3D by taking
two photographs
estimated by linear
regression from
gender-age-weight-height
0 to 3 y.o. 3 to 12 y.o.
15. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Children body shape space
• 800 registered 3D body scans of children
• Harmonised in posture
• PA + PCA to synthesise
• Shape & postural body descriptors
• Affordable no. of parameters
16. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Children body shape space
PC1
PC2
PC3
PC60
….
• 800 registered 3D body scans of children
• Harmonised in posture
• PA + PCA to synthesise
• Shape & postural body descriptors
• Affordable no. of parameters
17. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Input data
Image
processing
Iterative
optimisation
Data
extraction
How it works
18. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Introduction
Body measuring app
Size advice algorithms
Validation & results
Conclusions
Ongoing work
19. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Size advice algorithms
• Size recommended to wear it straight away → Expert’s advice
• Best fit to allow room for the child to grow → Parents’ advice
• Fit at different body areas → Traffic Lights
20. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Size advice algorithms
• Size recommended to wear it straight away → Expert’s advice
• Best fit to allow room for the child to grow → Parents’ advice
• Fit at different body areas → Traffic Lights
21. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Size advice algorithms
• Size recommended to wear it straight away → Expert’s advice
• Best fit to allow room for the child to grow → Parents’ advice
• Fit at different body areas → Traffic Lights
22. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
160 children aged 0-12 y.o.
Training data
NECK
OPENING
CHEST
WAIST CONTOUR
HIP
CONTOUR
SHOULDER
WIDTH
TOTAL LENGTH
ARMHOLE
BACK
WIDTH
Over 1100 fit trials
Key body fitting areas
WAIST CONTOUR
HIP CONTOUR
THIGH CONTOUR
FRONT RISE
KNEE CONTOUR
BACK RISE
TROUSERS
LENGTH
TROUSERS CUFF
23. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Ordinal Logistic Regression with Stepwise variable selection
𝑙𝑜𝑔𝑖𝑡 𝑃 𝑓𝑖𝑡 ≤ 𝐴|𝑋1, … , 𝑋𝑘 = ln
𝑃 𝑓𝑖𝑡 ≤ 𝐴|𝑋1, … , 𝑋𝑘
1 − 𝑃 𝑓𝑖𝑡 ≤ 𝐴|𝑋1, … , 𝑋𝑘
= 𝜃𝐴 | 𝐴+1 − 𝛽1 𝑋1 + ⋯ + 𝛽𝑘 𝑋𝑘 ,
𝐴 ∈ 𝑠𝑚𝑎𝑙𝑙, 𝑔𝑜𝑜𝑑, 𝑙𝑎𝑟𝑔𝑒 ∶ 𝑠𝑚𝑎𝑙𝑙 ≤ 𝑔𝑜𝑜𝑑 ≤ 𝑙𝑎𝑟𝑔𝑒
𝑃 𝑓𝑖𝑡 = 𝑠𝑚𝑎𝑙𝑙 | 𝑋1, … , 𝑋 𝑘 = 𝑒 𝜃 𝑠𝑚𝑎𝑙𝑙 | 𝑔𝑜𝑜𝑑 − 𝛽1 𝑋1+⋯+𝛽 𝑘 𝑋 𝑘
𝑃 𝑓𝑖𝑡 = 𝑔𝑜𝑜𝑑 | 𝑋1, … , 𝑋 𝑘 = 𝑒 𝜃 𝑔𝑜𝑜𝑑 | 𝑙𝑎𝑟𝑔𝑒 − 𝛽1 𝑋1+⋯+𝛽 𝑘 𝑋 𝑘 − 𝑃 𝑓𝑖𝑡 = 𝑠𝑚𝑎𝑙𝑙 | 𝑋1, … , 𝑋 𝑘
𝑃 𝑓𝑖𝑡 = 𝑙𝑎𝑟𝑔𝑒 | 𝑋1, … , 𝑋 𝑘 = 1 − 𝑃 𝑓𝑖𝑡 = 𝑠𝑚𝑎𝑙𝑙 | 𝑋1, … , 𝑋 𝑘 − 𝑃(𝑓𝑖𝑡 = 𝑔𝑜𝑜𝑑 | 𝑋1, … , 𝑋 𝑘)
Fit modelling
• Pick biggest size with good fit
• If between sizes → Pick larger
• If all big or all small → No size available
• Consistency expert-parent → Expert Parent
Decision rules for picking a size from probabilities
24. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Upper bodyLower body Full body
Stepwise
selection of
variables for
final models
Height Knee heightCervical height
Mid neck
girthChest
girth
Waist
girth
Hip
girth
Back armpits
contour
Height
Cervical height
Hip
girth
Belly
girth
Head girth
Wrist girth
Thigh girth
Chest
girth
Mid neck
girth
Fit modelling
Expert
advice
Parents
advice
Fit-by-
area
25. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Introduction
Body measuring app
Size advice algorithms
Validation & results
Conclusions
Ongoing work
26. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Databases
& APIs
Add-on & iFrame for
e-commerce
3D reconstruction engine
Mobile App
for parents
Size recommendation engine
FRONTENDBACKEND
Back-offices for
producers & e-retailers
Prototypes for testing
27. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Precision of kidsize app
• 6 3D printed children figurines in 1:10 scale
• Each figurine measured with the app 10 times
• Body measurements digitally obtained from 3D reconstructions
• Mean Absolute Difference (MAD) calculated for each measurement
𝑀𝐴𝐷 =
1
𝑛
1
𝑟 𝑖
2
𝑚 𝑠
𝑖
− 𝑚𝑡
𝑖
𝑟 𝑖
𝑡=𝑠+1
𝑟 𝑖−1
𝑠=1𝑖
30. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Accuracy of kidsize app
• 34 children aged 3-12 y.o.
• Each child was scanned with 3D scanner (Vitus XXL) and with the app
• Body measurements digitally obtained from 3D reconstructions
• Mean Absolute Difference (MAD) calculated for each measurement
𝑀𝐴𝐷 =
1
𝑛
𝑚 𝐾𝑖𝑑𝑠𝑖𝑧𝑒
𝑖
− 𝑚3𝐷𝑠𝑐𝑎𝑛
𝑖
𝑖
31. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Kidsize vs.
3Dscan
Expert vs.
Expert*
3Dscan vs.
Expert
Non-expert
vs. Expert
Cervical height 11 (1%) 7 8 50
Knee height 10 (3%) 6
Mid neck girth 11 (4%) 6 13 19
Neck base girth 11 (3%) 11 17 19
Shoulder length 14 (13%) 4 13 26
Chest girth 21 (3%) 15 19 24
Back Armpits Contour 20 (7%) 10 8 65
Waist girth 18 (3%) 11 20 24
Seat girth 12 (2%) 12 10 57
Arm length 13 (3%) 6 21 32
Outside leg length 13 (2%) 13 11
Thigh girth 14 (4%) 6 36
Knee girth 8 (3%) 4 18
Accuracy (MAD in mm)
32. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Kidsize vs.
3Dscan
Expert vs.
Expert*
3Dscan vs.
Expert
Non-expert
vs. Expert
Cervical height 11 (1%) 7 8 50
Knee height 10 (3%) 6
Mid neck girth 11 (4%) 6 13 19
Neck base girth 11 (3%) 11 17 19
Shoulder length 14 (13%) 4 13 26
Chest girth 21 (3%) 15 19 24
Back Armpits Contour 20 (7%) 10 8 65
Waist girth 18 (3%) 11 20 24
Seat girth 12 (2%) 12 10 57
Arm length 13 (3%) 6 21 32
Outside leg length 13 (2%) 13 11
Thigh girth 14 (4%) 6 36
Knee girth 8 (3%) 4 18
Accuracy (MAD in mm)
3D scan kidsize 3D scan kidsize 3D scan kidsize
3D scan kidsize 3D scan kidsize 3D scan kidsize
33. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Size advice testing
19 garments
Subjective assessment
Interview with retailers
‘Think aloud’ with parents
Objective assessment
Kidsize vs. parents’ choice
based on real try-on
Locations
Bóboli shop
IBV Lab
Participants
30 children
Aged 0-12 y.o.
(23 parents)
34. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Size advice reliability
Labelling
(age)
Size guide
(stature)
Kidsize
Expert 42% 54% 85%
Parents 48% 59% 88%
35. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Introduction
Body measuring app
Size advice algorithms
Validation & results
Conclusions
Ongoing work
36. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Conclusions
body
measuring
instrument
Size advice &
fit prediction
algorithms
37. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Conclusions
body
measuring
instrument
Size advice &
fit prediction
algorithms
38. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Conclusions
body
measuring
instrument
Size advice &
fit prediction
algorithms
39. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Introduction
Body measuring app
Size advice algorithms
Validation & results
Conclusions
Ongoing work
40. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Ongoing work & milestones
3D body reconstruction of adults
Robustness of segmentation algorithms
41. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Ongoing work & milestones
3D body reconstruction of adults
Robustness of segmentation algorithms
Improvement of 2D3D reconstruction
Validation of app with 200 adults
Health risk indicators
3D body reconstruction from RGB-D shots
Size advice for footwear
42. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
• FP7-SME-2013-606091. "Development of a new extended product-service to overcome size assignment and
fitting barriers for children fashion on-line market addressing customer needs" (KidSize), FP7, EC
• Alemany, et al. (2013). A Exploitation of 3D body databases to improve size selection on the apparel
industry, 4th Int Conf on 3D Body Scanning Technologies, Long Beach, CA, USA, November 2013
• Ballester et al. (2014). 3D-based resources fostering the analysis, use, and exploitation of available body
anthropometric data. 5th Int Conf on 3D Body Scanning Tech, Hometrica Consulting, Lugano, Switzerland
• Barrios et al. (2016). Reliability and criterion validity of self-measured waist, hip, and neck circumferences.
BMC Medical Research Methodology, 16(1)
• Bradtmiller & Gross (1999). 3D Whole Body Scans: Measurement Extraction Software Validation.
• Dekker (2000). 3D human body modelling from range data (Doctoral). University of London.
• Gordon et al. (1989). 1988 Anthropometric Survey of US Army Personnel-Methods and Summary Statistics
• Han et al. (2010). Comparative analysis of 3D body scan measurements and manual measurements of size
Korea adult females. International Journal of Industrial Ergonomics, 40(5), 530-540
• Lu & Wang (2010). The Evaluation of Scan-Derived Anthropometric Measurements. IEEE Transactions on
Instrumentation and Measurement, 59(8), 2048-2054
• Parrilla et al. (2015). Low-cost 3D foot scanner using a mobile app. Footwear Science, 7(sup1), S26–S28
• Robinette & Daanen (2006). Precision of the CAESAR scan-extracted measurements. Appl Erg, 37(3), 259-265
• Yoon & Radwin (1994). The accuracy of consumer-made body measurements for women’s mail-order
clothing. Human Factors: The Journal of the Human Factors and Ergonomics Society, 36(3), 557–568
References
43. 7th Int. Conference on Body Scanning Technologies
1st December 2016, Lugano, Switzerland
Alfredo Ballester
alfredo.ballester@ibv.upv.es
Ana Piérola
ana.pierola@ibv.upv.es
Sandra Alemany
Eduardo Parrilla
Jordi Uriel
Cristina Pérez
Paola Piqueras
Beatriz Nácher
Clara Solves
Julio Vivas
Silvia San Jerónimo
Juan C. González
anthropometry.ibv.org
www.kidsizesolution.com
Thank you !
Notes de l'éditeur
Kidsize provides size advice according to the brand (to be worn right now) and to parents (usually allowing for growth) as well as fit predictions.
Size recommendation is given as traffic lights with colour coded information. (Screenshots). Different areas considered for different garments, different areas are shown for different garments.
Fit predictions are especially important when you have to compromise on fit in order to buy the garment.
The two brands who participated in the project were interested in both types of information being given. Unsatisfied expectations.
Kidsize provides size advice according to the brand (to be worn right now) and to parents (usually allowing for growth) as well as fit predictions.
Size recommendation is given as traffic lights with colour coded information. (Screenshots). Different areas considered for different garments, different areas are shown for different garments.
Fit predictions are especially important when you have to compromise on fit in order to buy the garment.
The two brands who participated in the project were interested in both types of information being given. Unsatisfied expectations.
Kidsize provides size advice according to the brand (to be worn right now) and to parents (usually allowing for growth) as well as fit predictions.
Size recommendation is given as traffic lights with colour coded information. (Screenshots). Different areas considered for different garments, different areas are shown for different garments.
Fit predictions are especially important when you have to compromise on fit in order to buy the garment.
The two brands who participated in the project were interested in both types of information being given. Unsatisfied expectations.
Our algorithms were trained using experimental data from over 1100 fit trials.
11 garments of different types where provided in all their size spans by two childrenswear brands (French and Spanish).
160 children aged 0-12 y.o. and their parents voluntarily participated in the fit trials. Children were measured using traditional methods (0-3 y.o.) or using Vitus XXL 3D body scanner (3-12 y.o.).
At each fit trial, the size was evaluated by a fashion expert and by parents who declared if they would pick the size or not. Moreover, the expert also assessed the fit at the relevant body areas. These areas were defined were depending on the type of garment.
We chose to use Ordinal Logistic Regression Models with Stepwise variable selection. A total of 12 different models were defined according to: the source of advice (brand and parents), the age (babies 0-3 y.o. and children 3-12 y.o.) and the type of garment (upper body, lower body and full body). In addition two models (one for babies and one for children) were also defined for each of the 19 body fit areas (38 fit-by-area models in total).
The expert rules that applied to the output probabilities of the models were very simple:
The recommended size was the biggest size with good fit.
If the child was between two sizes (there is no size with a clear good fit), the recommended size was the bigger of the two.
If all sizes were big or all sizes were small, the system would not recommend any size.
The parents’ advice was always equal or one size bigger than the experts’.
19 and 36 body dimensions had been measured respectively for babies and children.
The original measurement sets were reduced to 9 variables by the stepwise variable selection (figure 6).
Expert size != parent size ... Probabilities are consistent
Colour coded answers. Child is between sizes
Every time the parent clicks on the Kidsize button, the child tries on all the sizes of the garment.
Only size with all ‘Good’ areas
Expert probabilities != parent probs although recommended size is the same
In order to assess the reliability of the size advices provided by Kidsize Solution, we made a test involving a group of volunteers meeting the target customer profile: parents with one or more children that sometimes buy childrenswear online. 30 children (23 parents) participated in the test. They were selected to have a gender- and age-balanced sample (10 children aged 0-2 years old, 10 children aged 3-7, and 10 children aged 8-12).
The childrenswear brands participating in the project (Bóboli and Sucre d’Orge) provided 19 different garments (figure 8) in all the available sizes for the testing.
The tests took place at two locations: at a Bóboli shop located in a mall (C.C. Bonaire) in Valencia (figure 9) and at a simulated multi-brand shop at IBV facilities. We prepared a fully working online demo shop including the Kidsize add-on button in the product selection page so that the participants could use it on a tablet during the tests (figure 9). Temporarily, Kidsize app was uploaded to Google Play in order to facilitate its distribution to the testers (figure 9).
During the test, each participant was asked to use the Kidsize app to measure his/her child and to go shopping four garments at the online demo shop. The four selected garments were randomly assigned to each participant according to a balanced design of experiments. For each of the garments evaluated, the parents tested first the size (or sizes) advised by Kidsize (expert’s and parents) and were encouraged to test other sizes. Then, the parents were requested to pick one size for wearing the garment straight way and to pick the size that they would buy (usually letting room for the child to grow); they were allowed to select the same size in both cases. At the end of the test, they were also requested to answer some questions about their experience using Kidsize.
The reliability of the size advice provided by Kidsize was compared with the available alternatives, i.e. a size guide based on children stature provided by the brands, and the labelling of garment sizes, which uses the ages as reference. Results are presented in Table 4.
Regarding the subjective evaluation, Kidsize solution was positively assessed by parents, who found it very promising for resolving their concerns when buying without trying on the garments. In general, it was considered very useful, reliable and easy-to-use. Some of the respondents requested a similar tool for adults.