Innovative AI initiative for second-hand textiles takes a step forward

Through an innovative collaboration, Wargön Innovation and RISE Research Institutes of Sweden have taken important steps towards developing AI-driven sorting and valuation of second-hand textiles. Already in 2021, the project AI for circular fashion - an early effort to explore AI solutions for the industry before the technology had its broad breakthrough. Now, this work has reached an important milestone that opens up for concrete applications of AI in the textile industry.

- We realized the potential of AI for the textile industry early on and have therefore been involved from the start of the great wave of development in AI that is now sweeping the world. The fact that we invested in this in 2021 has given us a good foundation for the future," says Susanne Eriksson, project manager at Wargön Innovation.

Comprehensive dataset for AI training

The target of 30 000 garments has now been reached. Each garment has been annotated with over 15 attributes and a front and back image, as well as a close-up of the brand. This dataset forms the basis for training AI models that can identify and value second-hand textiles based on image recognition. In addition to Wargön Innovation, Myrorna, one of Sweden's largest second-hand players, has been an important contributing force in the collection work.

The latest update of the dataset also includes a fraction with more detailed annotations on the condition of the garments, more specifically stains and holes. This provides the basis for AI models that can provide more detailed assessments of the value and usability of garments. Within the EU project CISUTACproject, which focuses on solving the bottlenecks in the European circular textile industry, value and usability have been identified as some of the most important valuation aspects.

Open resource for AI innovation

In line with Wargön Innovation's open philosophy, a first version of the dataset has already been released, freely available for research and development. As of today, the updated version will also be made open, enabling more actors to train and further develop AI models for sorting and valuing second-hand textiles.

- This is unique as it is the first large dataset used on garments from second-hand flows. It opens up for AI innovation in an industry where there is a huge need for efficiency and sustainability," says Susanne Eriksson, project manager at Wargön Innovation.

Concrete pilot projects underway

The dataset has already started to be used by RISE to train AI models that will be published in the fall. One of these models will be used in a pilot focusing on automated sorting within the EU project CISUTAC.

- "By applying AI to the sorting and valuation of second-hand textiles, we can significantly increase capacity and reduce the amount of clothing that currently goes to incineration or landfill. This is an important piece of the puzzle for a more circular textile industry," says Farrukh Nauman, AI researcher at RISE, who has been technically responsible.

With the comprehensive dataset in place, the future looks bright for AI-driven innovation in second hand and textile recycling - an industry where Wargön Innovation and its partners are positioning themselves as leading players.

The full dataset is open on Zenodo: https://zenodo.org/records/12518734

For more information please contact:
Susanne Eriksson, Project Manager AI for Circular Fashion, Wargön Innovation
Gabriella Engström, Project Manager CISUTAC, Wargön Innovation
Farrukh Nauman, AI expert at RISE and technical manager within AI for Circular Fashion