[2022-06-13] Le Twin, C’est Chic: Digital Twins for Fashion Application


Event Details

  • Date:
  • Categories:

Le Twin, C’est Chic: Digital Twins for Fashion Application

Marcelino Rodriguez Cancio
Couture Technologies

As e-commerce becomes a rising force in the fashion retail landscape, it faces a major, costly challenge: online purchases are subject to a much higher rate of returns (30-50%) compared to the rate for items purchased in-store (8.9%). These returns diminish profit margins, costing retailers money, time, and manpower. On the other hand, thousands of physical stores close all over the world each year, forcing companies to increase their online presence. Not only this is an economic problem; just in 2018, 5 billion pounds of clothing went to landfills due to returns, with the consequent environmental impact.

To address the online returns fashion e-commerce pain point, an exciting solution is the Virtual Try-On (VITON) technology, that promise to deliver custom fit and sizing visualizations to retail customers. While an intuitive idea, VITONs are full of technical risk. Cloth simulation speed and accuracy, anthropomorphic representation, and ease of use are areas requiring major improvement for a full-on implementation at the current scale of the fashion industry. 

This talk will address two major uses of digital twins in VITON: (i) anthropomorphic representation, and (ii) cloth simulation. We will also cover the solutions existing so far for accurate human and cloth modeling, as well as the challenges still remaining in the field.

 

Speaker bioThe people in my family who loves me the less call me the ‘malvestío’. Which in Spanish means the ‘poorly dressed’. Then, by an irony of destiny, I became the CTO of a Fashion Technology startup. Now, while I’m still a ‘malvestío’, my wife asks me to pick her outfits. My job consists of running a turf cyberwar for the control of Virtual Fitting City, alongside a gang of technological outlaws. During my Ph.D. two Benoits allowed me to tinker with the C++ code of the OpenJDK so we can make compiler optimizations that would allow numerical simulations and ML programs to run faster, at the expense of a bit of accuracy (it was a lot of fun). Before that, I got my Ms.C. in Digital Signal Processing by simulating the human blood flow for medical applications. 

Registration: https://jku.zoom.us/meeting/register/tJIvdeitqD8iGNNjcaltxNO0rmq1XaZBGUTB