Skip to content
Skip Start

Applied Research Impact Case 6

Using Artificial Intelligence to Improve Garment Size Fitting Predictions from 3D Body Scan and Psychographic Data

Dr Arthur Chan, Faculty of Design and Environment

 

When we think of Artificial Intelligence (AI), we tend to think of robots, social media, or facial recognition. AI can also be used in the apparel design process, for example for virtual fitting, drafting garment patterns for producing clothes, production management, and even to forecast retail sales. In online shopping, AI is important for enhancing customers’ shopping experience and enabling customers who are not physically present to select the “perfect fit” garment size for their body shape. To do this, an AI program needs to be trained to learn the relationship between garment fit and body dimensions.

 

Dr Arthur Chan and his team researched to develop an AI model that could enhance the accuracy of garment size and fitting predictions for the clothing industry. In the first stage of their study, the body dimensions of 50 participants, including men and women aged 25 to 39 years, were measured using a Skyku 3D body scanner (Figure 1). Since wearers’ fashion style, preferences, and lifestyle activities, known as “psychographic characteristics,” can influence garment size and fit, such information was also collected from the participants.

 

A1

Figure 1. Examples of participants’ 3D scan images

 

In the second stage, Dr Chan used a simulation software called Optitex to create a “virtual body” on the computer for each participant based on his/her scan measurements. A 3D T-shirt was then digitally projected onto the virtual body in a step known as “virtual fitting”, allowing the participants to see how the T-shirt would look and fit on the virtual self (Figure 2). Once completed, the 3D T-shirt could be flattened into 2D garment patterns, and the measurements were recorded for the third stage “AI analysis”.

 

A2

Figure 2. Fitting a virtual T-shirt onto a

virtual body using a 3D simulation software

 

For the AI analysis, Dr Chan entered the data into a program called “Artificial Neural Network” (ANN) to establish a size and fitting prediction model. The ANN consists of multiple layers of “neurons.” Like a human brain, ANN can cleverly learn the patterns between garment fit, body measurements and psychographic characteristics in a process called “Deep Learning” to create a size and fitting model. The study’s findings showed that the predicted size, fitting and measurements for garment patterns from the ANN were accurate. Figure 3 showed the predicted garment pattern measurements being plotted against the actual garment pattern measurements. In the diagram, the plotted data, i.e. “the dots”, clustered closely to and along the diagonal trend line. This indicated that the predicted results were very close to the actual values, thus suggesting a high level of accuracy of the ANN.

 

A3

Figure 3. Actual Pattern Measurements vs

Predicted Pattern Measurements Using ANN

 

Dr Chan conducted further analysis by segmenting the different body dimensions and participants’ psychographic characteristics, and it showed a significant improvement to the results. The predicted measurements and the actual measurements grouped even closer together on the trend line (Figure 4), meaning the difference between predicted and actual measurements were very small. Dr Chan concluded that the garment size prediction accuracy of ANN could be further improved when body dimensions and psychographic characteristics were taken into consideration.

 

A4

Figure 4. Actual Pattern Measurements vs

Calculated Pattern Measurements Using ANN by Clustering

 

The AI model developed in this project can be an ideal communication tool for manufacturers, retailers and consumers. A new understanding of sizing and fit could result in garments which fit consumers better, thereby reducing returns, improving consumer satisfaction, and increasing profit for retailers and manufacturers. It also contributes to sustainability by reducing manufacturing and consumption wastage. The project provides insight into developing a garment fit customisation database using AI technology. The AI models developed in the project are practical and can be used for automatic pattern generation in garment manufacturing.

 

(Acknowledgement: This project was supported by the THEi Seed Grant Scheme, Project No.: SG1819104.)