A story of innovation 


Only a so open-minded and innovative company could have seized the day, accepting the challenge of I4X in the AI application to image recognition, in order to improve its fabric cataloging system.

A historical archive of great value, which today has been completely revolutionized.


Company Name: Vicenza Mode

In the heart of Veneto, Vicenza Mode, a company that can boast a 100% Made in Italy production, has been working with passion in knitwear manufactoring for over 30 years. In order to mantain high quality standards, Vicenza Mode has been able to overcome market challenges innovating their business creating a virtuous bridge between the past and the future.

How we worked together

Our intent is to apply our Artificial Intelligence system to the recognition of the fabrics present in the company database. The AI ​​is in fact able to recognize the colors and geometries of any type of fabric.

The goal is to revolutionize the traditional industrial cataloging system through an advanced method: no more product codes, no more paper archives, each product is recognized and cataloged uniquely through images. Thanks to AI, technology and digital systems.

Not only that, in addition to knowing if the sample is already in the database, the information analyzed by the AI ​​adds other advantages, for example, knowing if that fabric is customizable.

Main results

The CNN network is based on the transfer training of AlexNet with the Matlab Deep Learning Toolbox and Parallel Computing Toolbox. The dataset has been obtained by acquiring four images for each fabric in four different poses: frontal, zoom, rotation clockwise and counterclockwise.
Using the rear camera of a Surface Pro 4 Tablet PC, NVIDIA® GeForce® RTX 2080 Super as the training hardware, 529 classes were trained in 67 minutes. The developed app is a standalone application obtained by Matlab Compiler.
The result is excellent even with unstructured light conditions, Surface Pro 4 in hand, with unstructured variable light, all the 529 fabrics are recognized.
By submitting non-trained fabrics, the network classifies those likely to be closest to the acquired image, sometimes surprisingly. If, for example, we submit to the net a motherboard of a PC, it proposes all the checkered fabrics.

Model-free approach

Our approach based on CNN do not needs any modelling assumptions

Would you like to experience a new approach in your company?