Capturing Textile Properties Using Deep Learning

In this project, 30 Omdena machine learning engineers partnered with Senstile, a startup in the textiles industry, to develop a search engine capable of analyzing an uploaded textile image and providing the best matches for alternative textiles. Senstile believes that such a system can dramatically improve the efficiency of the procurement process, which currently requires shipping physical samples to customers, as no other approach provides enough information about textile physical characteristics for customers to make informed decisions. 

The problem

Thousands of physical fabric samples are shipped daily in the fashion industry to make fashion happen. Designers, manufacturers, and suppliers must touch each fabric to understand the critical physical properties of the material. The developed AI system helps them identify similar textiles, substitute other materials, and optimize the selection for styles to be created.

The project outcomes

The specific objective of this project was to develop a search engine powered by algorithms that perform better than currently available technologies at analyzing a given textile sample and producing a selection of alternative textiles for customer consideration.  

To accomplish this objective, the team researched and implemented various pre-processing and modeling approaches, encompassing both traditional machine learning and deep learning techniques. Several of these efforts produced results worthy of further consideration and refinement. These options were incorporated into the search engine code developed and tested by Omdena using Streamlit and migrated to Senstile’s AWS instance for implementation.